Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

A pen-tip pressure-based system for real-time handprinted character recognition Willey, Stephen Raymond 1979

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-UBC_1979_A7 W54.pdf [ 12.71MB ]
Metadata
JSON: 831-1.0094674.json
JSON-LD: 831-1.0094674-ld.json
RDF/XML (Pretty): 831-1.0094674-rdf.xml
RDF/JSON: 831-1.0094674-rdf.json
Turtle: 831-1.0094674-turtle.txt
N-Triples: 831-1.0094674-rdf-ntriples.txt
Original Record: 831-1.0094674-source.json
Full Text
831-1.0094674-fulltext.txt
Citation
831-1.0094674.ris

Full Text

A PEN-TIP  P R E S S U R E - B A S E D SYSTEM FOR  R E A L - T I M E HANDPRINTED CHARACTER RECOGNITION  by  S t e p h e n Raymond B.Eng.,  Willey  McMaster U n i v e r s i t y ,  1976  A T H E S I S SUBMITTED I N P A R T I A L F U L F I L M E N T OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF A P P L I E D S C I E N C E in THE FACULTY OF GRADUATE STUDIES in  the  Department of  E l e c t r i c a l Engineering  We a c c e p t  this  the  thesis  required  as c o n f o r m i n g standard  THE U N I V E R S I T Y OF B R I T I S H COLUMBIA September, (cY  1979  S t e p h e n Raymond W i l l e y ,  1979  to  In p r e s e n t i n g  t h i s t h e s i s i n p a r t i a l f u l f i l m e n t o f the r e q u i r e m e n t s f o r  an advanced d e g r e e a t the U n i v e r s i t y o f B r i t i s h C o l u m b i a , I a g r e e t h a t the L i b r a r y s h a l l make i t f r e e l y a v a i l a b l e f o r r e f e r e n c e and s t u d y . I f u r t h e r agree that permission  f o r extensive copying  f o r s c h o l a r l y p u r p o s e s may be g r a n t e d by h i s r e p r e s e n t a t i v e s .  o f this thesis  by the Head o f my Department o r  I t i s understood that copying  or publication  o f t h i s t h e s i s f o r f i n a n c i a l g a i n s h a l l not be a l l o w e d w i t h o u t my written  permission.  Department o f  £l£crrticAt-  SMSW£££JNG  The U n i v e r s i t y o f B r i t i s h Columbia 2075 Wesbrook P l a c e V a n c o u v e r , Canada V6T 1W5 Date  II  OC^dllL  tfytj  ABSTRACT  An o n - l i n e h a n d p r i n t e d implemented,  and t e s t e d .  subset o f the dedicated  The s e l e c t e d  ANSI s t a n d a r d  graphics  an e n v i r o n m e n t  character  editor  system.  available  strain-gauge inferences  array.  a pressure-sensitive  of the paper.  motion.  a p p l i e d at  redundant aid  of  extraction is  a l l borders  It  is  placed  immediate not  scheme  total  software  a l s o v e r y CPU e f f i c i e n t  that s i g n i f i c a n t l y reduces  data.  In the  for  s i m p l e and e f f e c t i v e .  of the q u a n t i z a t i o n plane  final  stage input  structure  of l o c a l v a r i a t i o n s .  provision  monitored  w e r e p l a c e d on t h e w r i t i n g m e d i u m .  1500 1 6 - b i t w o r d s .  of a modified tree  variety  tendencies.  the  burden.  Feature is  provide  S i n c e p o s i t i o n a l i n f o r m a t i o n was  owing to a m u l t i - l e v e l p r e - p r o c e s s i n g computing  tip  to provide  The s y s t e m i s m i c r o p r o c e s s o r c o m p a t i b l e w i t h than  proposed  by p e r i o d i c a l l y s a m p l i n g a s t r a t e g i c a l l y  concerning stroke  comprising less  for a  F o r c e i n f o r m a t i o n was  D a t a were a n a l y z e d i n r e a l - t i m e  a v a i l a b l e no c o n s t r a i n t s  numerics  the n o n - t e c h n i c a l u s e r ' s n a t u r a l  i n the p l a n e  to the processor  i n c l u d e d the  A p r i m a r y d e s i g n o b j e c t i v e was t o  that would r e f l e c t  generated  symbol class  and n i n e s p e c i a l symbols d e f i n e d  A Xebec b a l l - p o i n t pen w i t h drag forces  r e c o g n i t i o n scheme w a s d e v e l o p e d ,  that  is  The d i c t i o n a r y i s  19 s y m b o l s r e q u i r e s  less  than  a l l a l l o w a b l e f o r m s n e e d be e x p l i c i t e l y  Extensive s m a l l amount  of user  t e s t i n g was c o n d u c t e d training is  required i i  to help  symbols are inherently  A hysteresis  scheme  eliminate  classified with  tolerant  to  a  large  compact and q u i c k l y  250 w o r d s  the  since only a  traversed; fraction  specified.  on t h e  completed r e c o g n i z e r .  to ensure adequate  A  performance.  A t o t a l o f more  t h a n 4000 c h a r a c t e r s  obtained  E x p e r i e n c e d u s e r s a c h i e v e d a 96.3% a v e r a g e 98% w e r e n o t position  uncommon.  Performance  and u s e r s a p p e a r t o r e t a i n  f r o m 15 s u b j e c t s  recognition rate  i s independent,of skills  s m a l l amount  o f hardware  Efforts to  were  a p a r t i c u l a r user.  reconfiguration v a l u e s were  character  results  I n some c a s e s ,  for a l l subjects.  some p a r a m e t e r to s e l e c t  fine-tuning.  and  acquired during t r a i n i n g .  d i r e c t e d to determine  robust,  size  above  Non-  distracted  by  involved.  improved performance  fairly  analyzed.  and s c o r e s  t e c h n i c a l and t e c h n i c a l u s e r s p e r f o r m e q u a l l y w e l l and a r e n o t the  were  i f the  s y s t e m c o u l d be  i t was found t h a t markedly.  dictionary  Optimal system  optimizing for a single subject E x c e p t i o n a l w r i t e r s , however, An automated  o p t i m a l v a l u e s was d e v e l o p e d and  i i i  parameter  y i e l d e d good  might b e n e f i t  personalization prototype tested.  dedicated  from  designed  TABLE OF CONTENTS  PAGE  ABSTRACT  .  i i  T A B L E OF CONTENTS  iv  L I S T OF TABLES  v i i  L I S T OF FIGURES AND I L L U S T R A T I O N S  viii  ACKNOWLEDGEMENTS  x  I.  INTRODUCTION  . . . .  . . . . . .  1  1.1  Preamble  1.2  General Concepts  1.3  Design Constraints  4  II.  A XEBEC PEN-BASED APPLICATION  7  2.1  G e n e r a l C o m p a r i s o n o f R e c o g n i t i o n Schemes  2.2  A X e b e c P e n - B a s e d Scheme  2.3  Character  III.  1  Set  and C o n s i d e r a t i o n s  for  . . . 1  .  7 9  a Dedicated Environment  10  THE B U I L D I N G BLOCKS OF CHARACTER RECOGNITION  .  .  .16  3.1  Data A c q u i s i t i o n  16  3.1.1  General Considerations  16  3.1.2  System C o n s t r a i n t s  17  3.1.2.1  Hardware  3.1.2.2  Software  .  . . . .  3.2  Pre-Processing  3.2.1  General Concepts  • .  .  17 18 21  .  .  iv  .  21  PAGE  3.2.2  A Multi-Level Pre-Processing  3.3  Feature Extraction  3.3.1  Purpose  3.3.2  Implementation  3.4  The C l a s s i f i c a t i o n Scheme  3.4.1  Introduction  3.4.2  D i c t i o n a r y Format  3.5  Testing  3.5.1  Spurious  3.5.2  Direction-Code  .  .  at  .  .  .  Scheme  .  21 28  .  .  .  .  28  Constraint?  28 .  .  .  .  .  .  .  .  IV.  .  36  and O p e r a t i o n  39  Intermediate Stages  Data at  Development  .  36  Stroke  . . .  End-Points  . . . . . . . . . . . .  Vector Calculation for  .  47 47  System Parameter  Dictionary Entry Optimization 3.5.3  .  and  .  49  of User T r a i n i n g - A i d s  50  F I N A L EVALUATION OF THE COMPLETED RECOGNITION SYSTEM  Dictionary  .  .  4.1  Classification with a Restricted  4.1.1  The L e a r n i n g C u r v e  4.1.2  Performance  4.1.3  Effect  4.1.4  R i g h t - and L e f t - H a n d e d n e s s  63  4.1.5  Training for  65  4.1.6  Confusion Matrices  4.1.7  Skill  4.1.8  C l a s s i f i c a t i o n Robustness  4.2  Construction  4.2.1  Introduction  4.2.2  Reconfiguration  of Character  to User Category  S i z e on P e r f o r m a n c e  95% R e c o g n i t i o n  .  •  • •  51 53  Sensitivity  Retention  .  51  „ .  .  .  .  .  .  .  .  f o r N o n - n o m i n a l Forms  of P e r s o n a l i z e d  Systems  56 60  67 70 72 74 74  of the  C l a s s i f i c a t i o n Tree  v  75  PAGE  4.2.3  Threshold Factors  4.2.4  Sample-Rate  4.2.5  C l i p p i n g Time  81  4.2.6  Time-Out  83  4.2.7  Plane Non-Uniformity  4.2.8  A u t o m a t i n g the  4.3  Performance  4.3.1  Effects  4.3.2  R e c o g n i t i o n Performance  V.  • » - 78  Considerations  79  Value  84  P e r s o n a l i z a t i o n Process  w i t h an E x t e n d e d R e s t r i c t e d  upon R e c o g n i t i o n o f t h e for the  . . .  87  Dictionary  O r i g i n a l Character Extended Symbol Set  . . . . . . Set  88  . . . .  88  . . . . . .  93  CONCLUDING REMARKS  96  5.1  Summary o f E x p e r i m e n t a l R e s u l t s  5.2  Achieved Design Objectives  5.2.1  Favorable  Characteristics  5.2.2  E l i m i n a t i o n of Undesirable  .  . . . . .  i  . . .  . . . . . . . . .  101  . . . . . . . .  101  Characteristics  5.3  System L i m i t a t i o n s  5.4  Comparison w i t h P u b l i s h e d R e s u l t s  5.5  Recommendations  5.6  Conclusions  for  96  .  106  Further  of Other Authors  107  Research  no .  .  116  REFERENCES  APPENDIX  104  118  A  Example o f R e c o g n i t i o n System Performance  B  Example of D i s t o r t e d C h a r a c t e r  C  Photograph  of Character  Forms  R e c o g n i t i o n Hardware  vi  . '.  123 124 125  L I S T OF TABLES  TABLE  I  PAGE  System C h a r a c t e r i s t i c t s Directly  Influenced  Drawn f r o m P u b l i s h e d L i t e r a t u r e  that  Design Objectives  Effects  III  Average R e c o g n i t i o n Performance  54  IV  Performance  A f t e r 15 M i n u t e s o f T r a i n i n g  57  V  ANOVA T a b l e  for Users  VI  Mean a n d V a r i a n c e o f D i f f e r e n c e  VII  C l a s s i c a l Hypothesis  Testing  VIII  Confusion Matrix for  R e c o g n i t i o n Performance  IX  "6"  from D i f f e r e n t  25  Categories  . . . .  59  Data  62  .  . . . . .  62  After  - Large D i g i t s .  Confusion Matrix Training  on D i g i t  5  II  Training  of Pre-Processing  . . .  -  68  f o r R e c o g n i t i o n Performance  After  Small Digits  •  of D i c t i o n a r y Reconfiguration  68  X  The E f f e c t  XI  Summary o f P a r a m e t e r S e n s i t i v i t y i n a F r e e F o r m E n v i r o n m e n t  XII  Summary o f A N S I V e r s u s E x t e n d e d D i c t i o n a r y T e s t R e s u l t s  92  XIII  Hypothesis  Testing:  Varying  the  Dictionary  93  XIV  Hypothesis  Testing:  V a r y i n g the  Symbol Set  vii  76  .  .  .  .  . . .  78  94  L I S T OF FIGURES AND ILLUSTRATIONS  FIGURE  PAGE  1.1  The B u i l d i n g B l o c k s  2.1  Example o f a T y p i c a l D i s p l a y U s i n g the Electronic  2.2  3.1  Circuit  of Pattern  .  . . . . . .  Set  Draw i n P a r t  . . . . .  2  Computer-Assisted  Program  Extended Gesture (b)  Recognition  Configured f o r the  Demonstration  .  .  12  .  .  14  Circuit .  P l o t o f I n i t i a l a n d Maximum D e v i a t i o n o f X - , Y - a n d Z-Pressure  Values  19  3.2  Flowchart of the  3.3  A G r a p h o f Raw X - P r e s s u r e D a t a f o r Character,  Recognition System's  Pre-Processing  the  A Derived Topological Plot  3.5  Example o f the  Effect  for  . . . .  22  . . . . . . . . . . .  24  Handprinted  "P"  3.4  Stage  the Handprinted  Character,  of Adding Small Hysteresis  Zones  "P"  . .  30  at  Quantized Direction-Zone Borders  32  3.6  Summary o f Q u a d r a n t  3.7  S e v e r a l Examples o f T r a i n i n g Program Output  3.8  Examples Demonstrating  3.9  Example o f C l a s s i f i c a t i o n Tree S t r u c t u r e  3.10  Examples o f Stroke E n d - P o i n t Data  43  3.11  An I l l u s t r a t i o n  of Inherent  45  3.12  A Demonstration  of the  Variations  Q u a n t i z a t i o n Schemes  the  Effect  . . . . . . . .  .  .  .  33 .  .  .  .  .  .  .  of M u l t i - L e v e l P r e - P r o c e s s i n g : .  . • • • • »  . . . . . .  D i c t i o n a r y Compactness  D i c t i o n a r y ' s Tolerance  -  37 /$!  to Symbol  . o f the  35  46  3.13  Structure  Linked-List  C l a s s i f i c a t i o n Scheme  4.1  Example of a Completed R e c o g n i t i o n Performance  4.2  A L e a r n i n g Curve of S i n g l e T r i a l  viii  S c o r i n g Sheet  R e c o g n i t i o n Performance  48 . . .  52 55  FIGURE  4.3  PAGE  Testing Attained  4.4  Data  f o r S u b j e c t s whose R e c o g n i t i o n  Performance  95% . . . . . . . . .  The E f f e c t  66  o f Time F o l l o w i n g T r a i n i n g on R e c o g n i t i o n  Performance  .71  4.5  Recognition  Performance w i t h Non-Nominal Input  4.6  Examples  4.7  The E f f e c t  of Changing P r e - S e t P r e s s u r e Thresholds  4.8  The E f f e c t  o f V a r y i n g Sample  4.9  The E f f e c t  o f V a r y i n g Time-Out V a l u e and P l a n e  4.10  Specimens o f S u b j e c t s '  4.11  Example o f D i s p l a y e d Output  of Variants  ^ - 73  from N o m i n a l ANSI Forms  77  R a t e a n d C l i p p i n g Time  . . .  .  Configuration  I and I I H a n d w r i t i n g from t h e Automated  Scheme 4.12  Testing with  an E x t e n d e d  Restricted  5.1(a)  Differentiating  the Z e r o ,  5.1(b)  Differentiating  t h e S i x and N i n e  5.2  Negative  Oersted,  of Hysteresis  ix  .  .  . 82  .  .  . 85 86  Personalizaton  .  Effects  * • 80  89 Dictionary and S i x . . . . . .  91 108 108 112  ACKNOWLEDGEMENT  I Drs. In  would l i k e  M.R. Ito  provided  my s i n c e r e  and C F . S c h r a c k f o r t h e i r  a d d i t i o n , I wish  helpful  to express  t o e x t e n d my t h a n k s  appreciation  constant to other  interest  to and  f a c u l t y members  encouragement. for  their  d i s c u s s i o n and t o V . W o n g , V . L e i , T . L e u g n e r a n d D . D a i n e s who assistance  i n the  i m p l e m e n t a t i o n phase of  am g r a t e f u l  t o t h e many f a c u l t y ,  outside  department  the  Financial Scholarship  fellow  graduate  this  thesis.  students,  who p a r t i c i p a t e d i n s y s t e m t e s t i n g  assistance  was p r o v i d e d t h r o u g h a  o f f e r e d b y the N a t i o n a l R e s e a r c h C o u n c i l  x  Finally,  and f r i e n d s experiments.  Postgraduate of Canada.  I  from  I,  1.1  INTRODUCTION  Preamble  The a b i l i t y essential  to  interfaces oriented  the  i n the  user  to  creative field  i n mind.  f u n c t i o n i n g were  store,  often  thought  process.  Until  T h o s e who p o s s e s s e d required  Success  resulted  sensitivity  and the  programmer's  information  recently,  from the  fine  the  technicallymachine  to i n t e r - d i s c i p l i n a r y  i n t e r a c t i o n of the  astuteness.  Reports  artist's  describing  and m u s i c composers  the  appear i n  the  [FRAN 7 8 ] . The q u a l i t y o f i n t e r a c t i o n depends h e a v i l y upon a v a i l a b l e  output  devices.  Their s u i t a b i l i t y to  great importance. through  the  In this  thesis,  (e.g.  microcomputer)  user's natural  1.2  pattern  a pen-paper  constraints environment  application is  interface  character  of real-time that  is  scheme.  recognition i n the  of  explored  recognition  reflects  input/  a  non-technical  tendencies.  G e n e r a l Concepts and  In  the p a r t i c u l a r  implementation of a handprinted  E m p h a s i s i s p l a c e d upon t h e limited  is  man-machine  minimal knowledge of  to apply themselves  of writers, graphic a r t i s t s  literature  and t r a n s f e r  o f computing were d e s i g n e d w i t h  endeavors.  struggle  manipulate  F i g . 1.1  recognition  Considerations  below are  system.  depicted  the major b u i l d i n g b l o c k s o f  a  2 Low level  I/P  High level  Preprocessing  Data Aquisition  ;  Defined c o l l e c t i v e l y as the A n a l y s i s stage  Origin of a message or d i r e c t i v e Fig.1.1  The B u i l d i n g B l o c k s  of Pattern  The r e c o g n i t i o n p r o c e s s the  is  description of acquired data.  often  a b i n a r y d i g i t stream;  fulfillment  of a d i r e c t i v e .  of  definite  rules  It  is  at  general,  its  the  Consequence o f a message or d i r e c t i v e  Recognition  an a t t e m p t  In i t s  to progress!vely  lowest  highest  form, the  level i t  At intermediate  and c r i t e r i a ,  t h i s procedural process In  Classification  Feature Extraction  stages,  information i s  that defines  a r e c o g n i t i o n system w i l l  "message"  may d e f i n e through  the  application developed.  recognition  adhere t o  two  is  the  increasingly  a pattern  improve  scheme.  basic  principles: a)  a l l v a l i d inputs repeatable  fashion.  always generate process b)  is  the  a g i v e n message w i l l  Tou quantities  defines  of input  Striking  the  results  letter  in  a  " H " , for example,  same m a c h i n e r e s p o n s e .  In this  sense  should the  unique;  no t w o i n p u t s  is  must p r o v i d e p r e d i c t a b l e  always correspond  generate the  the  categorization  into various pre-defined  a c c o m p l i s h e d by e x t r a c t i n g  input  and  same c o m m u n i c a t i o n .  r e c o g n i t i o n as  data  to a p a r t i c u l a r  only those  of  classes  features which  large  [TOU 7 2 ] .  appear  This  significant  from a background o f i r r e l e v a n t d i r e c t i o n d u r i n g the K a n a l has  course  detail.  of the  certain strong  a p r a c t i c a l system.  He e x p r e s s e s  of  approach  the  theorem-proof  there i s  a growing tendency  simplicity,  the need  features  is  the  pragmatic  conducted to v e r i f y  regarding  concern over the  to pattern  conception  apparent h i g h He p o i n t s  [KANA 74].  "experimentation  to determine  The d e v e l o p m e n t  i n approach.  that  appear on bank cheques.  symbol d i s c r i m i n a t i o n i s procedure  is  simplified.  At  the  printed postal are be  o f the  Where p o s s i b l e ,  and  other  end o f the  code r e c o g n i t i o n  extracted  must be  the  class  flexible  [GENC 6 8 ] .  experiments  the  In this  f o r example,  system.  Constraints  potential variations  the  characters  classification  problem of  hand-  environment the  thesis  the  relates  various  character  are  infinite.  to  classified? Features a  character  characters.  c l o s e l y to  on s y m b o l f o r m a t i o n a r e m i n i m a l ,  f o r each  characters  s y m b o l "4-" i s  a n d "4" t h e n b e  s u f f i c i e n t l y s p e c i f i c t o d i s t i n g u i s h amongst  latter  simplest  enough t o a l l o w v a r i a t i o n s w i t h i n  i n this  the  author's  of numeral four or numeral nine?  Research undertaken  of  spectrum l i e s  r e c o g n i z e d a s n u m e r a l f o u r , how s h o u l d belong to  and  only i n nominal form,  As a r e s u l t ,  If,  that  suppositions.  Since they e x i s t  facilitated.  out  suitable  r e c o g n i t i o n of magnetic i n k  almost e n t i r e l y unconstrained.  Does "V  yet  the  thi  of  Elsewhere i n  From an i m p l e m e n t a t i o n p o i n t o f v i e w p e r h a p s the common a p p l i c a t i o n c o n c e r n s  in  emphasis  i n t h e name o f p r a c t i c a l i t y  studies  assumptions  the  recognition.  only worthwhile route".  s y s t e m was f a i r l y were  feelings  to question,  that  was e x p e n d e d  thesis.  for theoretical  same a r t i c l e h e a d v i s e s  A major e f f o r t  the  the number  The A N S I  standard  numeric set  of ten d i g i t s  comprise the  1.3  [ A N S I 74]  Design  Constraints  l i s t e d characteristics of s e v e r a l  a c c o m p l i s h e d schemes t h a t author's system.  results in  due  testing  Unfortunately,  to a lack of consistent procedures.  Character  it  sets,  alphabet  restricted  and unambiguous  repertoire  Table I provides  is  term  and t o  more  design  compare  of  published  inconsistencies  to the  the fairly  u s e d b y H o s k i n g [HOSK 7 2 ] .  dictated  (as  to both  favorable  design decisions are  and  for  the  self-explanatory,  opposed to  'static')  a word processing  given.  techniques  since  the  numeric " b " .  If  r e c o g n i t i o n schemes r e l y on s t a t i c  scanning of input  sequence i n f o r m a t i o n .  C o n s i d e r the  they were  off-line  two s y m b o l s , a l p h a b e t i c  analyzed, post  chronological ordering of  strokes,  it  facto, is  without  unlikely  A d y n a m i c scheme w o u l d h a v e  in  them s i n c e  d i s c r i m i n a t i n g between the  holds  letter  for  the  "b" yet  ambiguous  time  " h " and  any k n o w l e d g e o f  that  they w o u l d be  little  difficulty  "o" was made i n a c l o c k w i s e  a n t i - c l o c k w i s e for  potentially  processing  does n o t y i e l d  readily distinguishable.  for  the  range from  [CASE 6 4 , STAL 75]  'dynamic'  Optical character  the  that  data  to  f o r example,  A l t h o u g h m o s t comments  of e x p l a n a t i o n of the  difficult  a quick reference  characteristics  system.  is  testing  c o m p l e x Chinese  undesirable  of the  d i r e c t l y or i n d i r e c t l y i n f l u e n c e d  relatively  proposed  symbols  repertoire.  I n Table I are  the  together w i t h nine dedicated,  pair,  fashion  digit  A similar  argument  letter  " S " and d i g i t " S " .  5 TABLE I . A  SYSTEM CHARACTERISTICS DRAWN FROM PUBLISHED L I T E R A T U R E THAT D I R E C T L Y INFLUENCED DESIGN  A U T H O R [REFERENCE]  Hosking  Groner  Bernstein [BERN 6 8 ]  Miller [MILL  69]  Ledeen [LEDE  73]  Chui [CHUI 76]  FAVORABLE  UNDESIRABLE  1) R e l i a n c e upon a particularly unambiguous c h a r a c t e r set  2) D a t a p r o c e s s e d on a 2) H i g h - s p e e d c o m p u t i n g point-by-point basis f a c i l i t i e s required reducing intermediate 3) Many s t y l i s t i c s t o r a g e and i m p r o v i n g c o n s t r a i n t s imposed response time  [GRON 6 8 ]  [BERN 6 9 ]  C H A R A C T E R I S T I C S  1) R e a l - t i m e s y s t e m benefits (e.g. dynamic p r o c e s s i n g )  [HOSK 7 2 ]  Bernstein  OBJECTIVES  & Howell  3) R e c o g n i t i o n p e r f o r m ance independent o f character size  4)  4) E l i m i n a t i o n o f spurious end-point data before feature extraction begins  5) E x e s s i v e l y h i g h memory requirement  6) R e c o g n i t i o n P e r f o r m ance i n d e p e n d e n t o f overall character s l a n t and p o s i t i o n  6) C h a r a c t e r s i z e c o n s t r a i n e d ( t o y-")  Choice o f strokes r a t h e r than characters as s y s t e m p r i m i t i v e s . This necessitates determining relations between s t r o k e s .  7) A n a l y s i s i s n o t b e g u n u n t i l stroke construct i o n i s completed  7) E m p h a s i z e s i m p l i f i c a - 8) U s e r s a r e r e q u i r e d t o e n t e r e a c h member o f t i o n of pattern the c h a r a c t e r s e t analysis thereby several times before reducing computing r e c o g n i t i o n may b e g i n burden 8) P e r f o r m a n c e i n s e n s i t i v i t y to l o c a l variations.  9) U s e o f a n e n v i r o n m e n t which i s l e s s n a t u r a l t h a n pen and p a p e r  TABLE  1/8  SYSTEM CHARACTERISTICS DRAWIN FROM PUBLISHED L I T E R A T U R E THAT D I R E C T L Y INFLUENCED DESIGN O B J E C T I V E S ,  A U T H O R  C H A R A C T E R I S T I C S  [REFERENCE]  FAVORABLE  Chui Cont'd [CHUI 76]  9) M i n i m a l  Crane & S a v o i e  10)  [CRAN  77]  CONT'D  user  training  Micro-computer compatibility  UNDESIRABLE -  10)  R e s u l t s were not proved s t a t i s t i c a l l y significant  11) M i n i m a l preprocessing requiring a large number o f d i c t i o n a r y references 12)  Specific error rates under v a r y i n g c o n d i t i o n s were not published  7 II.  2.1  A XEBEC P E N - B A S E D  G e n e r a l C o m p a r i s o n o f R e c o g n i t i o n Schemes  T h e common k e y b o a r d t e r m i n a l i s input  APPLICATION  devices. (a)  It  does,  however, have s e v e r a l  the majority o f people be u n d e s i r a b l e  (b)  to teach  t h e i r primary  use  of a non-conventional character  pads might prove  computer  may  input  is  task;  keyboards.  set  requires  The use o f  confusing to n o n - t e c h n i c a l  manual r e t r a n s c r i p t i o n entails  type and i t  them e s p e c i a l l y i f d a t a  not  used  drawbacks:  do n o t know how t o  (and u s u a l l y expensive)  (c)  one o f t h e m o s t w i d e l y  function  users;  ( o r i g i n a t o r to paper  a d d i t i o n a l overhead  special  to  i n terms of c o s t ,  keyboard) delays  and  errors. U s i n g a pen d e v i c e would e l i m i n a t e (a)  handprinting is  (b)  accomodation of non-standard  an a l r e a d y e s t a b l i s h e d  through software the (c)  these d i f f i c u l t i e s  characters  skill  since:  f o r most  is easily  users;  effected  m o d i f i c a t i o n s which remain transparent  to  user;  the need  f o r h i g h overhead  retranscription  is  completely  obviated; (d)  the  pen generates i t s  for  future  printing  reference.  Using a keyboard requires  a  stored  seperate  device.  One m i g h t s u s p e c t is  own o r i g i n a l c o p y w h i c h may b e  e x c e s s i v e l y slow but  c o n t r o l l e d experiments  that  the  Devoe p r o v e s  information throughput o t h e r w i s e by c i t i n g  w h i c h show t h a t  the  for  results  speed o f h a n d p r i n t i n g  handwriting from numerals  8  and c a p i t a l s typist  compare  [DEV067].  interspersed instruction  favourably with  the  k e y p u n c h i n g r a t e o f an  This statement i s p a r t i c u l a r l y v a l i d i f input  with other represent  tasks.  Computer-aided design or  ideal potential  the  it  is  also a natural  design objective  that  the  is  computer-assisted  situations.  N o t o n l y d o e s h a n d p r i n t i n g p r o v i d e an e f f i c i e n t transfer,  unskilled  form o f i n p u t .  It  means o f  fulfills,  e n v i r o n m e n t be s u i t e d  at  data  least  to comfortable  in  part,  human  interaction. In  the  past  optical character line.  There  a major e f f o r t  recognition  are numerous  o f r e s e a r c h e r s has been d i r e c t e d  (OCR) s c h e m e s  inherent  i n which data  disadvantages  are  towards  processed  i n using this  form o f  processing: (a)  time-sequence  o r d y n a m i c i n f o r m a t i o n w h i c h c o u l d be u s e f u l  discriminating various digit (b)  the at  "5")  i s not  p o s s i b i l i t y of error the  OCR e q u i p m e n t  (d)  r e c o g n i t i o n performance  deterioration,  An a u t o m a t e d  (e.g.  letter  and subsequent  significantly  r e l a t i v e l y expensive  image q u a l i t y and i s  (a)  detection  time of w r i t i n g i s is  pairs  " S " and  available;  (c)  ink  character  is  reduced;  and  space-consuming;  d i r e c t l y dependent upon s y m b o l ,  therefore  sensitive  smudge m a r k s , d i r t ,  pen-based  correction  system that  to paper  and  etc.  functions  on-line  exhibits:  advantages a c c r u i n g to dynamic p r o c e s s i n g such  as:  i)  information,  character e.g., stroke  d i s c r i m i n a t i o n v i a time-sequence  a one-stroke digit  in  "5",  letter  "S" d i f f e r s  from the  two-  off-  ii)  visual  or audio feedback  a l l o w i n g immediate  of m i s c l a s s i f i e d or u n c l a s s i f i e d (b)  a p o t e n t i a l l y low c a p i t a l investment  (c)  i n s e n s i t i v i t y to following  the  defining  The scheme  does n o t  listed  2.2  suffer  or scanners  They can a l s o u n r e a s o n a b l y  small writing  devices. that  .  surfaces  of the  hard-copy  and t e n d t o  constrain  the  or by i n c o r p o r a t i n g u n n a t u r a l  d e s c r i b e d i n the n e x t s e c t i o n i s t h e s e drawbacks y e t  be  user  by  writing  a pen-based  does e x h i b i t a l l o f t h e  system advantages  above.  A X e b e c P e n - B a s e d Scheme  X e b e c S y s t e m s ' A l p h a b e c 75 p r e s s u r e together  with  purchased found  the  associated  as a r e s e a r c h  to be  fairly  analogue  tool.  incompatible with  was  not  p e n i s moved i n t h e  of the  define  plane  character  An a r r a y to a f l e x i b l e forces.  was  b o a r d was a l s o s u p p l i e d  the h o s t  processor.  Inserting  gates r e c t i f i e d the problem.  but a  Software  p e n and A / D m o d u l e a p p e a r s a s A p p e n d i x C . o f the  a p p l i e d by the w r i t e r are  reliably  (A/D). c o n v e r t e r  acquired.  A photograph  those  s e n s i t i v e b a l l - p o i n t pen  to d i g i t a l  An i n t e r f a c e  number o f c o n t r o l l i n e s and l o g i c  the  and s p a c e - e f f i c i e n c y ;  formation.  Many p e n s y s t e m s e m p l o y t a b l e t s e x p e n s i v e and b u l k y  data;  c o n d i t i o n or treatment  character  correction  paper,  generated  at  forces the  e q u a l and o p p o s i t e  pen-tip.  These drag  As  to forces  shape.  of photographically etched s t r a i n  diaphragm detect  As the pen i s b r o u g h t  instantaneous i n t o contact  gauges  that  v e r t i c a l and l a t e r a l with  are  bonded  pen-tip  the w r i t i n g medium,  the  diaphragm i s m i n u t e l y deformed. digitized of  the  and s u b s e q u e n t l y  compound b r i d g e  components  are  "pen-up"  information character  pressures schemes  transmitted  o u t l i n e d i n the  (i.e., often  literature  v e r t i c a l p r e s s u r e component  P e n movement  i s i n f e r r e d through  that  pen-tip  reflect  their  the  input  processsor  pen p o s i t i o n , c o n s t r a i n t s size.  condition.  analysis  is possible This type  of  to  detect  dynamic  ambiguous  of instantaneous  d i r e c t i o n of t r a v e l . data  to determine  i s not  Many  drag  position-based  relative  necessary  are n e i t h e r  directions  when u s i n g a  pressure-  d e f i n i t i o n for  Character  receive  any i n f o r m a t i o n  regarding  p l a c e d on t h e w r i t i n g m e d i u m n o r  Set  for  A numerics was u s e d i n e a r l y  characters.  The n e x t s e c t i o n d e a l s  the proposed  on  had i n d i c a t e d t h a t p o s i t i o n  a n d t h a t a t i m e s e q u e n c e o f p e n d i r e c t i o n s was  t o p o s i t i v e l y i d e n t i f y most character-set  does not  Early experimental results  was n o t n e c e s s a r y  recognition  data  sufficient with  system.  a Dedicated Environment  only symbol set  testing.  conforming to  to the  s e t was d e f i n e d .  a repertoire  t h a t would have immediate  University of B r i t i s h  the  R e c o g n i t i o n performance  an e x t e n s i o n  the  force  pen. Since  2.3  it  in discriminating certain potentially  [CHUT 7 2 , M I L L 6 9 , POWE 7 3 ] ; t h i s  character  Details  pairs.  pre-process  tracking  three-dimensional  are  [CRAN77].  o f f t h e w r i t i n g medium)  aids  orthogonal strains  to processing algorithms.  used i n i s o l a t i n g the  By s a m p l i n g t h e the  The a s s o c i a t e d  Rather  than  p r o p o s e d ANSI  standard  was s o e n c o u r a g i n g  select  symbols  at  that  random,  a p p l i c a t i o n i n a scheme e x i s t i n g  C o l u m b i a was c o n f i g u r e d .  A l t h o u g h the  p e n was  at not  11 implemented  i n the  a p p l i c a t i o n were  chosen environment,  the  important  Under the  o r i g i n a l implementation  device i s  used  (refer  various  s y s t e m commands  to F i g . 2.1  an a b s o l u t e - p o s i t i o n  for t y p i c a l output).  are  referenced  over defined  s u b - a r e a s and t h e n  defined  (termed  the  "menu")  assembly,  is  to  the  right  of the  screen  the keyboard,  to  i n extending  , ( f l u i d i t y o f p e n movement)  s y m b o l a m b i g u i t y was possible,  t o be  inter-character  (c)  extraneous or unnecessary  (d)  diagonal strokes rotated  were  this (e)  avoided  was a  desirable  ( m a x i m i z i n g , where  discrimination); stroke  segments were n o t  avoided since  under  quadrant quantization plane  r e c o g n i t i o n performance  i s sensitive  the  (refer to  allowed;  selected  to S e c t i o n  3.3)  local variations  in  region;  c h a r a c t e r s s h o u l d be e a s i l y u n d e r s t o o d ( T r i e s m a n and Kahneman s y m b o l s " s h o u l d be  advised that  a prime  and  the  remembered  use  of  "familiar  consideration).  _ Professors British  The and  the  feature; (b)  and  cross-hairs  on t h e k e y b o a r d .  adhered  dictionary: w r i t i n g ease  refresh  d i s p l a y e d menu.  C e r t a i n s y m b o l d e s i g n g u i d e l i n e s were  (a)  on a Megatek  p o s i t i o n i n g the  The p e n w o u l d r e p l a c e and the  design.  E l e c t r o n i c components  by f i r s t  located  circuit  joystick/cross-hair  s t r i k i n g any c h a r a c t e r  always p r e s e n t .  joystick/cross-hair  current  electronic  to i d e n t i f y or p o s i t i o n graphical objects  display  is necessarily  an  examined.  The a p p l i c a t i o n i n v o l v e s c o m p u t e r - a i d e d  area  aspects o f such  T r i e s m a n a n d Kahnemen a r e  members  C o l u m b i a ' s Psychology department  o f the  [Private  Univeristy of  communication].  the  13 Because  of  special character more  than  merely  expression.  the  expressed  c o u l d be r e p l a c e d  The g e s t u r e s e t  plates  are  In the configured to  delete  it"  from r i g h t  f o r the  hard  anticipated  to  copy.  left,  model demonstration  To r e v e r s e  (b)  program.  score  of  components  100% w a s  The b l o c k i s with  the  hard-copy  the  a line  shown and i n p u t  "delete"  to  the  gesture  does n o t  an  arrow  h-parameter  referenced  d u r i n g the  That  recognition  activate  and " e n d " d i r e c t i v e s  a support/referencing  o f the  order  through  draws  the  as w e l l  session  and a  as  is,  program  circuit  design  six  recognition  o f system hardware  shows t h e  A / D c i r c u i t r y i n a s h i e l d e d box and a p r i s m - s h a p e d  i s moved i n t h e  strain  "draws  in  recorded.  pen f u n c t i o n s  magnitude  F i g . 2;2,  c r e a t e d when t h e  as  d e v i c e w h i c h when u s e d i n  an i n c r e m e n t a l  cursor.  machined c i r c u l a r o p e n i n g and a p p l y i n g a f o r c e cursor  to  meaningfully  to normal w r i t i n g d i r e c t i o n .  shows t h e  were  two  inductor.  s h o u l d be  user merely  for  a capacitor,  an e l e c t r o l y t i c c a p a c i t o r h e  In Appendix C, a photograph complementary  an  Again i n reference  r e c o g n i t i o n responses but  The " r e v e r s e " ,  electronic  to i d e n t i f y  commands  the  term  means o f  c i r c u i t was drawn i n a s i m u l a t e d e x p e r i m e n t .  c i r c u i t was d r a w n as  which outputs  user.  or  the  f o r e x a m p l e , was d e s i g n e d  In order  identified,  contrary  2.2  a natural  A coiled wire signifies  same w a y , d i r e c t i v e s  Figure  the  drawn.  (e)  r i c h e r word " g e s t u r e " w h i c h i m p l i e s  shown i n F i g . 2 . 2 ,  background.  a symbol already  on h i s  by t h e  of c r i t e r i o n  a memorized symbol; a g e s t u r e i s  a u s e r w i t h an e l e c t r o n i c s parallel  importance  desired  d i r e c t i o n at  applied force.  gauges, p o i n t s  on t h e  Because screen  the  pen-tip,  a rate proportional  of the  metal  block.  conjunction  By d i p p i n g t h e  at  Xebec p e n ,  pen i n t o  the  a displayed to  the  high resolution possible with  can be r e f e r e n c e d  accurately.  Since  14  (a)  Extended  Gesture  S i m u l a t i o n  Set  C i r c u i t  C o n f i g u r e d  Drawn  i n  f o r  P a r t  the  (B)  GESTURE  COMPONENT OR DIRECTIVE  ~AA-  R e s i s t o r C a p a c i t o r I n d u c t o r F i e l d  E f f e c t  T r a n s i s t o r  V o l t a g e  Source  C u r r e n t  Source  D e l e t e  G r a p h i c a l  ®  Object  F i n i s h  (b)  /  E OR F  End  S i m u l a t i o n Gestures  —-AA-  C i r c u i t  I d e n t i f i e d  I n v o l v i n g  J  i  < - = ! =  +  J !  <v> T  i i  Most  of  the  Above  , f  [Reverse]  (Rightgoing)  4r  Reverse  r  ~  1  l  1  I  1  &  i  ' !  1 i  J  Delete  <£ i  I  i i  .JL  I  E [END] F i g u r e  2.2  15 the host processor,  a N o v a 840 m i n i c o m p u t e r , a c c o m o d a t e s  cursor p o s i t i o n i n g tasks Under the pre-defined  origin  anticipation  and a s e t  user,  the  on t h e  subsequent  pen t i p  the  tube.  raises  (a v i s u a l  pen-tip  s i m p l e and n a t u r a l  feedback  The  S i m i l a r l y , when  forces  are  in the  for the  user)  pressure until  the  connecting  gesture-defined  input,  generated  some  processor,  displays a capacitor with  19 s y m b o l r e p e r t o i r e  on s e v e r a l p a r t i c i p a n t s .  for example,  Subsequent  the pen.  r e q u i r i n g component r e f e r e n c i n g  referencing  If,  at  referenced  displayed cursor/cross-hairs  cursor,  on t h e M e g a t e k r e f r e s h  The c o m p l e t e tested  positioning task.  a n u l l v e r t i c a l pressure,  d i s p l a c e d by subsequent  c u r s o r would appear  or a cursor would appear.  shifts  s a m p l i n g , most  second.  s c r e e n e a c h t i m e a component i s  h a v i n g p o s i t i o n e d the  directives is  one  proposed implementation the  of cross-hairs  at  detecting wiring  than  "II", was d r a w n , t h e word c a p a c i t o r  gesture,  applied  o f the  require less  fast:  the  cursor appears  by the  user.  (including cursor  They a l l f o u n d component and and c u r s o r p o s i t i o n i n g  fast  and  control)  was  directive  and  accurate.  16 III.  3.1 3.1.1  THE BUILDING BLOCKS OF CHARACTER RECOGNITION  Data A c q u i s i t i o n General Considerations  Data a c q u i s i t i o n i s the conversion of information from i t s raw state into a form suitable f o r input to subsequent analysis phases. Imperfections at t h i s stage are d i r e c t l y r e f l e c t e d i n recognition performance. In the past, hardware to perform this task has involved one o r more of sensors (e.g. the Rand tablet [BERN68, BERN69, GRON66, GRON68, MILL69]), scanners and various forms of d i g i t i z e r s [POWE70, POWE73] a l l  implemented  with some degree of success. From a human-factors standpoint the universally accepted pen and paper environment would provide an excellent i n t e r f a c e .  Previous attempts  at applying s i m i l a r p r i n c i p l e s to the computing environment were unsuccessful for a number of reasons.  In some schemes, f o r example, users were forced  to look away from the paper at CRT when writing [GR0N68, MILL69].  In  others, non-marking o r cumbersome pens were employed o f t e n r e s t r i c t i n g written information to the confines of " t a b l e t " surfaces [CHUI76].  Efforts  have recently been placed on improving the aesthetic appeal of the i n t e r face [EXPE78]. sive.  Unfortunately these devices are often p r o h i b i t i v e l y expen-  Xebec Corporation manufactures a pen whose c h a r a c t e r i s t i c s closely  approximate the requirements c i t e d above.  The hardware i s presently quite  expensive since the item i s not commercially a v a i l a b l e .  However, because  of the system's inherent s i m p l i c i t y , a d r a s t i c reduction i n p r i c e could be r e a l i z e d i f the unit were mass-produced.  17 3.1.2  System  3.1.2.1  Constraints  Hardware  The N o v a 8 4 0 f u n c t i o n i n g u n d e r D a t a G e n e r a l ' s O p e r a t i n g S y s t e m and e q u i p p e d w i t h provided the  far  entire  character/gesture  5 milliseconds. i t was n o t In  at  c l o c k was a d j u s t e d  to  final  the  during i n i t i a l  (e.g.  refresh  disk files  stages o f the  d u r i n g the  course  implementation. a scheme  LED d i s p l a y )  was o b s e r v e d  values  at  the  respectively  tended  s y s t e m was  raw d a t a were  f o r subsequent  design.  the  The f i n a l for visual  a n d an  that in  a  of the  liked.  thesis,  they  analysis.  transferred A line-printer  raw d a t a m o n i t o r i n g and a p o i n t -  do n o t  system envisaged feedback  every  operational  Although these p e r i p h e r a l s  inexpensive  component  (each u n i t  as one m i g h t h a v e values  of  that i n the  output  service  d i s p l a y p r o v i d e d a g r a p h i c a l means o f v e r i f y i n g  Each o f the It  felt:  testing.  e a r l y stages o f system d e s i g n ,  tape or i n t o  electronics,  request i n t e r r u p t  T h i s r a t e p r o v e d e x c e s s i v e once the objectionable  intermediate  useful  is  r e c o g n i t i o n system could perform adequately  and a s t o r a g e - t u b e d e v i c e a l l o w e d r e a l - t i m e addressable  it  memory  environment.  A real-time  to magnetic  32k 1 6 - b i t words o f random a c c e s s  more p r o c e s s i n g p o w e r t h a n w a s a c t u a l l y r e q u i r e d ;  microcomputer  but  Real-Time Disk  o f the  serve  information  were  any p u r p o s e  comprises  the  in  pen,  the  i t s A/D  classification decision  microcomputer.  p r e s s u r e s was i n t e r r o g a t e d p e r i o d i c a l l y .  "pen-up"  c o n d i t i o n mean x - , y - ,  A/D converter  corresponds  to  Furthermore,  t o vary from c h a r a c t e r  and  were - 2 0 , - 1 2 and - 4 8  15/2  volts).  units  They were n o t  i t was d i s c o v e r e d  sample  z-pressure  to c h a r a c t e r  that  these  sample.  zero offset  18 A n i n d i r e c t s o l u t i o n was a d o p t e d observation:  although  "offset"  appreciable,  variations  fluctuations  over the  o f each  following  f o r m a t i o n time were n o t  as o f f s e t  to n o r m a l i z e a l l subsequent sample v a l u e s . stage s i m p l i f i e d  o f the  over a l o n g time i n t e r v a l were  pen-up  this  vector  character  a result  The f i r s t  problem at  force  as  character  was t h e r e f o r e  pre-processing  (Fig.3.1).  saved  and  Dealing with  and f e a t u r e  used  the  extraction  calculations. Applying this initially in  requires  h e l d f r o m t h e w r i t i n g m e d i u m . None o f t h e  complying to  3.1.2.2  scheme, i m p l i c i t l y  the  imposed  that  the p e n - t i p  subjects  felt  handicapped  constraint.  Software  Hardware i s NOVA a s s e m b l y  s u p p o r t e d by d a t a - a c q u i s i t i o n software  language.  Other p o r t i o n s  o f the  written  s y s t e m p r o g r a m were  F o r t r a n t o m i n i m i z e c o d i n g t i m e and t o f a c i l i t a t e m o d i f i c a t i o n s .  of  the  more s a l i e n t  features  o f system programs  Electrical  E n g i n e e r i n g at  o f the  are the  software  by a n a l y s i s  approach,  reduced intermediate  as  opposed  maintained  to  The n e c e s s i t y  acquired data  a reasonable  emptied  followed  level  and  need  for  requirements.  arises  in itself  from the  dictate  requirement  the  that the  n o t e x c e e d 15 m i l l i s e c o n d s e v e n t h o u g h  of  an a v e r a g e  requires  and  analysis.  acquisition  a c q u i s i t i o n p e r i o d must each sample  of  Columbia.  a sequential time at  Several  Further  various stages of pattern  response  data b u f f e r i n g  Department  was f i l l e d w i t h  The c o n c u r r e n c y a s p e c t does n o t circular buffer.  the  University of British  thus s y n c h r o n i z i n g the  Concurrency o f p r o c e s s i n g ,  discussed here.  a v a i l a b l e through  A 200-word c i r c u l a r b u f f e r simultaneously,  are  in  written  in  details  be  o f 40 m i l l i s e c o n d s .  the  W i t h o u t the  a  optimal  analysis circular  OBSERVATION CONCLUSION  N  2  -50  I n t e r - C h a r a c t e r Pressure F l u c t a t i o n s are Large Relative to Intra-Character Fluctuations. I n i t i a l x-, y-, and z-"Pen-Up" Pressures are R e p r e s e n t a t i v e o f Successive Values w i t h i n Any C h a r a c t e r Formation S e s s i o n .  I  5  X  I  o  -t—» O  OJ  >  X  -20 •  I  l_ </)  I  cu  Y  -10  X  2  3  4  5  6  7  Session N u m b e r '3:' denotes two i d e n t i c a l v a l u e s f o r a given s e s s i o n . I denotes v a l u e s which a r e w i t h i n one u n i t o f one another. F i g u r e 3.1  P l o t of I n i t i a l  and Maximum D e v i a t i o n Pressure Values  (Resting  State)  VO  20 buffer,  a n a l y s i s w o u l d be i n t e r r u p t e d b y t h e  current  d a t a were f u l l y  processed.  Incidentally,  the  generated  character  15 m i l l i s e c o n d v a l u e w a s n o t s e l e c t e d a r b i t r a r i l y .  features  a great d e a l of redundant  than  15 m i l l i s e c o n d s p r o v e d  went u n d e t e c t e d . data  author  the  ware d e r i v e d m u l t i p l e o f the  5ms. i n t e r f a c e  and s e v e r a l  r e s u l t i n g object  Under these c o n d i t i o n s the test  subjects  Dictionaries a i d of support  recognition 4.2,  code.  are  written  programs and are  session.  I n the  devices, short t o accommodate  i n assembly  soft-  assumes  handcould code.  classification  stage are  response.  generated  loaded i n t o core p r i o r to  with  each  " P e r s o n a l i z e d " approach d e t a i l e d i n S e c t i o n with pertinent:  t o t h e s y s t e m upon u s e r  optimal  parameter  identification.  the Xebec pen n o r the Megatek d i s p l a y p r o c e s s o r initialization  and d e v i c e h a n d l i n g r o u t i n e s  were  them.  a n a l y s i s , the next control.  A l l  t i m e was a p p r o x i m a t e l y 1 s e c o n d .  W i t h d a t a a c q u i s i t i o n c o m p l e t e and i n f o r m a t i o n f o r m a t t e d  processing,  The  The more  40 m s . v a l u e  complained of e x c e s s i v e l y slow system  automatically input  subsequent  a  a t t e m p t s w e r e made t o  Fortran routines  system response  used i n the  Since neither standard  c o l l e a g u e s and i s  A r e d u c t i o n from the  the most s u i t a b l e d i c t i o n a r y t o g e t h e r  values  of writers.  coded i n t i m e - e f f i c i e n t assembly l a n g u a g e .  o n l y have been a c h i e v e d by r e w r i t i n g the  the  period  clock.  c o m p l i c a t e d a l g o r i t h m s were coded i n F o r t r a n b u t  None o f t h e  fastest  unacceptable  a n a l y s i s p e r i o d o f 40 m i l l i s e c o n d s was e q u a l l y c o n s t r a i n e d .  h e a v i l y used r o u t i n e s were  optimize the  Reducing the  f o r even the  c h o s e n v a l u e was o p t i m a l f o r  The  the  .  U s i n g an a c q u i s i t i o n p e r i o d o f g r e a t e r since important  a c q u i s i t i o n task before  stage i n the  classification  scheme,  for pre-  are  21 3.2  Pre-Processing  3.2.1  General  Concepts  Pre-processing usually refers applied  to  available to,  reduce at  the  inherent  the  ultimately  determines  preparation  f o r the  is  c l a s s i f i e d as In  data  scheme; hence however,  final  salient  to  the need i n the  the  i s often  loss of important  referred  e x t r a c t i o n stages  i n which  the  unknown  and  sample  class. only significant recognition  Over p r e - p r o c e s s i n g  character  features,  could,  whereas  t o an: i n t o l e r a b l e  level™  chosen p r e - p r o c e s s i n g  scheme  section,  made,  o r g a n i z i n g them e f f i c i e n t l y i n  for pre-processing.  following  raw d a t a  to subsequent stages o f the  A f l o w - c h a r t o u t l i n i n g the I n the  it  o v e r a l l computing burden  too l i t t l e might degrade system response  as F i g . 3 . 2 .  feature  decision-making process  reduce  in. the  a n a l y s i s , as  and the  features,  allowed to progress  result  and d i s t o r t i o n s  Pattern  pre-processing the  some f o r m o f i n f o r m a t i o n , m a n i p u l a t i o n  a member o f a p r e - d e f i n e d  order  s h o u l d be  redundancies  acquisition stage.  comprises both  to  the  entire  process  is  applying  appears  described  in  detail.  3.2.2  A M u l t i - L e v e l P r e - P r o c e s s i n g Scheme  P r e - p r o c e s s i n g comprises reduction. quantities (i) (ii) (iii) (iv)  I n most h a n d p r i n t e d of irrelevant writing hand  data  surface  two a s p e c t s :  character result  s m o o t h i n g and  recognition systems,  data considerable  from:  inconsistencies;  tremor/slippage;  electronic noise quantization  and  effects  (position discreteness,  A/D conversion).  22 Pre-processing,  ^  ("interrupt c l o... c k i s software .driven q u a n t i t y of data xs [jreduced by a f a c t o r of 3 no  Verify i f user has started writing  D i s c a r d —^ F e t c h data next values data point  yes  © no  ©  C l i p 3 Data from Both End-points of A l l Strokes.  similar  significantly  different  Feature Extraction Stage (with Hysteresis)  F i g u r e 3.2  Flowchart  of the P r e - P r o c e s s i n g  Stage.  1 I  i i  Local data v a r i a t i o n s  can o f t e n  smoothing algorithms In the  [GRON56].  F i g . 3.3  are  Xebec pen d u r i n g the typical  plotted  the  sequence o f x - p r e s s u r e s  handprinting of letter  is  fairly  is  r e l a t i v e l y smooth and f r e e (a)  be e l i m i n a t e d b y a p p l y i n g p o i n t - a v e r a g i n g  o f numerous  pressure plots  "P".  The r e s u l t i n g  made f o r v a r i o u s  from the p e r t u r b a t i o n s  a hard tablet  effects  or d i g i t i z e r surface)  of inconsistencies  and  (as  by  graph  symbols.  mentioned  t h e w r i t i n g medium was a common p a d o f p a p e r to  generated  above  It  since:  opposed  therefore  and p o s i t i o n d i s c r e t e n e s s  were  negligible; (b)  the  pen i t s e l f  weighs o n l y 6 ounces  and i s  fairly  easy  to  use; (c)  a l l e l e c t r o n i c equipment  (d)  temperature  effects;  finally,  inherent  the  gauge a r r a y It  was  mechanical i n e r t i a of the  probably contributes  to  from  strain-  smoothing the  data. and t h a t  t i m e w o u l d be a p p l i e d to d e t e r m i n i n g w h e t h e r  data  are  data  reduction stage  of  s h i e l d e d and p r o t e c t e d  d e c i d e d t h a t s m o o t h i n g w o u l d n o t be n e c e s s a r y  gained computation  significant to  sequentially  is  executed  the  c l a s s i f i c a t i o n process.  of p r e - p r o c e s s i n g . levels  and the  F i g . 3.2  The a l g o r i t h m i s  quantity  o f data  is  the  incoming  describes  composed o f r e d u c e d by a  the five facto  8. Level  were  asked  character  1 deals w i t h v e r t i c a l ( i . e .  t o a l l o w the formation,  pen-tip  to  2) p r e s s u r e .  Although users  touch the w r i t i n g s u r f a c e  a number o f t h e m h a d  the  tendency  to  only  during  touch the  paper  1200 F i g u r e 3.3  A Graph o f Raw x-pressure Data f o r the Handprinted C h a r a c t e r , "P".  800'N c  ZD  400  0) t_  z> <J)  C D i_  O.  0  i  X  <-t—  50  100  ^0  Time (msec)  o  -400  c  -800  -1200  S3  before  moving the pen or t o  completion.  As a r e s u l t ,  r e c o g n i t i o n performance v e r t i c a l pressure  the pen i n t r a i l i n g  spurious stroke  By d e f i n i n g  t h r e s h o l d t h a t had to be exceeded b e f o r e  having tested  c o u l d n o r m a l l y be  subjects  tolerated.  eliminated p o t e n t i a l l y s i g n i f i c a n t data. generated  following  a minimum saved,  A numerical  to determine  Higher  symbol  and  d a t a were  p a t t e r n s were g e n e r a l l y d i s c a r d e d .  selected after  maximum t h r e s h o l d t h a t  verification,data  contact  d i r e c t i o n s were i n f e r r e d  suffered accordingly.  these spurious low pressure v a l u e was f i n a l l y  leave  the  values  As a b y - p r o d u c t o f Z - t h r e s h o l d  through a c c i d e n t a l p e n - t i p  contact were  also  eliminated. In of  a c o n t r o l l e d experiment,  raw d a t a g e n e r a t e d  d i g i t s i x were s t o r e d on a m a g n e t i c d i s k .  pre-processed  under  the  was o b s e r v e d to d e c r e a s e further  results  are  summarized i n Table I I  TABLE  PRE-PROCESSING  CRITERION  5%  subsequently  t o t a l number o f  f r o m 57 t o 5 4 .  This  data and  and i l l u s t r a t e d i n F i g . 3 . 8 .  II  E F F E C T S OF P R E - P R O C E S S I N G  SPECIMEN  These d a t a were  Z - t h r e s h o l d c r i t e r i o n and t h e by a p p r o x i m a t e l y  d u r i n g the p r i n t i n g  ON D I G I T " 6 "  % OF TOTAL NUMBER OF P O I N T S DISCARDED INDIVIDUAL  CUMULATIVE  A  none  0.0  0.0  B  Z-threshold  5.3  5.3  C  X,Y,Z-threshold  14.0  19.3  D  X , Y , Z - t h r e s h o l d and end-point clipping  24.6  43.9  26  The s e c o n d l e v e l I t was n o t e d  that  contact  the paper before  with  exceeded  subjects  of data  r e d u c t i o n concerns  would very often starting  allow the p e n - t i p  h i g h to-exceed  fairly  This  minimal.  fact  i s not  the pen.  X - and Y - t h r e s h o l d s  augmented  to  i n c l u d e the  a d d i t i o n to the  improved,  threshold,  drag pressures  surprising since  were t h e r e f o r e  Z - t h r e s h o l d ) had a l s o  often.  strokes.  that although v e r t i c a l pressure  the  remained  u s e r was n o t m o v i n g  d e f i n e d and  c r i t e r i o n that either  e s p e c i a l l y f o r the  fairly  the  pre-processing  the X - o r Y - t h r e s h o l d  t o be e x c e e d e d .  Performance  i n e x p e r i e n c e d u s e r who c o m m i t t e d t h e  A s an a d d e d a d v a n t a g e t h e  frequency  a further  above t e c h n i q u e t o  the  raw d a t a  5 d a t a p o i n t s were e l i m i n a t e d ( c f . Unintentional end-point  characters.  dictionary  non-significant  data e a r l y i n the process  this  3 is  end,  truncated  level  ( c l i p p e d ) by t h r e e d a t a p o i n t s  augments the decreasing results  likelihood  this  i n Chapter  otherwise  it  either  six  end.  the l e v e l ' s e f f e c t i v e n e s s .  by  the to  remove  time s u f f e r s .  Strokes Greater  are truncation  whereas  An a n a l y s i s o f  X - , Y - , Z - t h r e s h o l d v a l u e s and c l i p p i n g IV.  digit  is preferred  response  data e l i m i n a t o r . at  reduced.  handprinted  tolerated  of s i g n i f i c a n t data being discarded  value reduces  u s i n g extreme  presented  a simple end-point  o f most  forms are  (discussed i n Section 3.4),  end-point  ).  d a t a were a f e a t u r e  A l t h o u g h some o f t h e s e s p u r i o u s  design of the  c o l l e c t e d for the  Table I I  unintentional  of spurious  " h o o k s " a n d " t a i l s " p r o d u c e d b y more e x p e r i e n c e d w r i t e r s was a l s o By a p p l y i n g the  in  affected.  raw d a t a r e v e a l e d  was s u f f i c i e n t l y  pressures.  to r e s t  and d r a g f o r c e s were i n f e r r e d as m e a n i n g f u l c h a r a c t e r  A n a l y z i n g the  error  (x and y )  t o w r i t e ; t h e v e r t i c a l t h r e s h o l d , was  R e c o g n i t i o n p e r f o r m a n c e was a d v e r s e l y  (in  drag  factors  is  To  27  L e v e l s 1,2 fulfill later  a secondary  stages.  function  is  and 3 a r e p r i m a r i l y d a t a r o l e of reducing the  Level 4 is  a true  refinement  techniques  which  quantity of data transmitted  d a t a r e d u c t i o n t e c h n i q ue i n t h a t  t o unburden the system o f redundant  data.  to  its  major  Under schemes  that  f u n c t i o n w i t h p o s i t i o n a l i n f o r m a t i o n , a common d a t a - t h i n n i n g t e c h n i q u e discard  those  previously  data points which  accepted p o i n t .  w o u l d be t o accept lateral)  the  pressures,  significant. pressures  as  data  U s i n g the Xebec p e n , i f the  change i n e i t h e r  D a t a w o u l d be d i s c a r d e d i f the  were l e s s  X - and Y - p r e s s u r e  the  d i s p l a c e d from  corresponding algorithm of the  t h a n some p r e - d e f i n e d arises  changes  drag  (i.e., was  i n both. X - and Y -  threshold.  i n that there e x i s t s  cases where  e x c u r s i o n s are not i n d i c a t i v e of p e n - t i p  For example, drag pressure  changes  at  line  large  direction  segment  end-points  may  b e a p p r e c i a b l e e v e n t h o u g h new d i r e c t i o n i n f o r m a t i o n i s n o t g e n e r a t e d . exhibiting large saved is  only  also  changes  i f the  i n b o t h X - and Y - p r e s s u r e  r a t i o of pressures,  substantially different Consequently,  i.e.  should therefore  small  pressure  ratio At  data.  changes is  to a test  r e l a t i v e l y unimportant.  s i g n i f i c a n t are  this point,  But before  salient  describing this  reduction, which i s briefly.  are  applied after  data passed features  of drag pressure  are  the  feature  Only i f the  to subsequent selected  change  ratios. extraction, in  stages of a n a l y s i s .  from t h e  stage i n d e t a i l , a f i n a l feature  be  ratio.  Since i n f e r r e d d i r e c t i o n s are very c o a r s e l y q u a n t i z e d at stage,  Data-  the i m p l i e d s t r o k e d i r e c t i o n ,  from the p r e v i o u s  Level 4 resolves  to  the  compared to t h e p r e v i o u s l y a c c e p t e d p r e s s u r e s ,  A slight difficulty  change.  are not s u f f i c i e n t l y  is  level  remaining of  data  e x t r a c t i o n , s h a l l be e x p l a i n e d  Since p o s i t i o n a l i n f o r m a t i o n i s not Xebec d e v i c e ,  several  consecutive  data points  directly available each  d i r e c t i o n - c o d e p r o v i d e no more i n f o r m a t i o n t h a n direction-code.  Successive equal  L e v e l 5 i d e n t i f i e d i n F i g . 3.2 code  information.  3.3  Feature  3.3.1  sample  enough  are  same  of  that  be s a f e l y  these redundancies  discarded.  in  direction-  Extraction  the  Feature  computed.  Extraction stage,  The t e c h n i q u e  and r a p i d d i s c r i m i n a t i o n at to  tolerate unintentional In  Topological  the  past,  feature  sections,  were  inflections. technical  able  to extract  literature  of the  construction  the  c l a s s i f i c a t i o n stage,  character  geometric  feature  yet  to  allow  general  applied  to  the  problem.  the  defined  the  character  into  cusps,  extraction  eight  curvature  are noted  in  and the  CHUI76].  generally  take the  appearance o f a s t r i n g  string is  a statement  grammar.  S p e c i f i c grammatical  o f t h e s e s t a t e m e n t s and are  about the  considered  sample rules  i n the  in.  of  the  govern  following  the section.  Constraints  Powers concludes cues,  the  variations.  information regarding  [LEDE73, LIN74,  Implementation  positional  of  a p p l i e d must be s p e c i f i c enough  the parameter around  Features extracted  language  features  e x t r a c t i o n was e f f e c t i v e l y u s e d b y T o u a n d G o n z a l e z  Examples o f  p r i m i t i v e elements;  salient  a v a r i e t y o f methods have been  [TOU72] w h o , b y d i v i d i n g  3.3.2  eliminates  the  a single point  c o d e s may t h e r e f o r e  the  Purpose  At  exact  generating  from  "nearly  that short-term  characterizes  pen d i r e c t i o n i n f o r m a t i o n  natural  characters"  [P0WE73].  without The  implication  that  r e c o g n i t i o n performance  i n f o r m a t i o n were s u p p l i e d i s "one  reason  precision  substantiated  f o r u s i n g more t h a n  cannot  c o u l d be i m p r o v e d i f  b y F r e e m a n who c l a i m s  one p r o p e r t y  frequently  o t h e r w i s e be a c h i e v e d " [FREE61].  extracting other  further  than d i r e c t i o n a l features  is  that  that, the  The p o s s i b i l i t y  from d a t a  generated  desired of  by the Xebec  pen was e x a m i n e d . Early experimental strain  results  gauges c o u l d be r e s o l v e d t o s t r o k e  sampled at provides  a constant  estimates  of pen-tip  position.  vertical  and t o p o l o g i c a l g r a p h s  or absolute  as F i g . 3 . 4 ;  its  dissimilarity  The c o n s t a n t generated  who w r i t e f a i r l y  was f i n a l l y  Samples were o b t a i n e d  prepared  to  absolute  the  A plot  are implicit!;  relations  were the  from a v a r i e d  assuming constant  h o r i z o n t a l or  for character  drawn c h a r a c t e r  magnitude o f v e l o c i t y  worthwhile results, But i n g e n e r a l ,  to person  d e c i d e d to  velocity  pen's  contradicts  P appears the  horizontal velocity.  slowly.  from person  Since pressures  Several velocity  magnitude o f v e l o c i t y .  assumption o f constant  occasionally  direction.  from the  c a l c u l a t i o n s w o u l d be c o n s t r a i n e d b y  proposed microcomputer environment. population  data read  r a t e any a s s u m p t i o n r e g a r d i n g p e n - t i p  examined b e a r i n g i n mind t h a t  varying  indicated that  shift  and t e s t i n g the  emphasis  assumption  e s p e c i a l l y i n the f i n d i n g s were  discouraging,  session to t e s t i n g to optimizing  case o f c h i l d r e n  session.  the only  It  information  made d i r e c t l y a v a i l a b l e , p e n d i r e c t i o n . As i n d i c a t e d e a r l i e r resolution  force  non-nominal,  i.e.  common, f e a t u r e s  data. local  the  strain-gauge  array  generates h i g h  Since handprinted  character  v a r i a t i o n s i n the  form o f u n i n t e n t i o n a l d a t a  e x t r a c t e d have  t o be f a i r l y  coarse  forms are  i n nature.  inherently are  A small  Start  Y  A  Figure  3.4  A Derived  T o p o l o g i c a l P l o t f o r Handprinted Character,  Under the Assumption of Constant H o r i z o n t a l Speed.  "P",  31  slanted  digit  b e l o n g i n g to at  its  two, for example, w i t h t h e same p a t t e r n  head.  difficulty;  A human r e a d e r the  In segmented  an e x t e n d e d t a i l  class  as  a very large  s y s t e m s h o u l d do n o  Miller  important  topological  plane  ( i . e . , d i v i d e d i n t o f o u r e q u a l segments)  direction-code  features.  [GRON66].  but  other  as  the  for the  the modified  author's  system.  defined hysteresis  code remains unchanged.  fall  outside  d i r e c t i o n code be  the  are  for  describe  a uniform  a l l  quadrant the  generated  number complex  was n o t e d : force  plane i n  their  diagonal  lines  direction  fell  rotated  uniform quadrant  plane  As l o n g as i n f e r r e d d i r e c t i o n zones,  the  Assume, f o r example,  extracted 16° o f  hysteresis.  " 1 " , successive  ± ( 4 5 + 8 ) ° e x t e n d e d segment  quantized  i n order  coordinates that  a  generated.  The p r o p o s e d f e a t u r e hysteresis  results  d i d not l i m i t  instantaneous  a p r e v i o u s d i r e c t i o n was q u a n t i z e d to q u a d r a n t  different  [MILL69J.  of the 4 5 ° b o r d e r s .  3.5 i l l u s t r a t e s  remain w i t h i n  would have t o  for  uniform quadrant  A single difficulty  v a r i a b l e d i r e c t i o n codes  w h i c h was s e l e c t e d  If  unacceptable  Complex s y m b o l s , t h e r e f o r e ,  [CRAN77].  f r o m one s i d e t o t h e  direction  equal  statements.  r e c o g n i t i o n scheme  coordinates  quantized i n a  direction-codes  Groner opted  Crane and S a v o i e employed a r o t a t e d  Fig.  "hook"  little  codes were i n s u f f i c i e n t to a d e q u a t e l y  their  d i r e c t i o n codes  are  divided his plane i n t o eight  t o Xebec pen d a t a y i e l d e d  complex symbols s i n c e f i v e  as  less.  and d e f i n e d e a c h s y m b o l by f i v e  Applying such a technique  generated  two w i t h a s m a l l  the proposed system, h i g h r e s o l u t i o n data  direction plane.  classified  would c o r r e c t l y recognize both w i t h  classification  direction-segments  of  s h o u l d be  extraction algorithms with  summarized i n F i g . 3 . 6 .  and w i t h o u t  The a d d i t i o n a l o v e r h e a d  that  results  32 I„ Without  II.  Hysteresis  With 16° o f H y s t e r e s i s  D i r e c t i o n Codes : 1,2,1,2,*-  The S i n g l e D i r e c t i o n Code,  1,1,2,2 a r e Generated  1, i s Generated.  C h r o n o l o g i c a l l y f o r the  a t i o n Under Scheme I I i s  Nominally H o r i z o n t a l B-stroke  Simpler and F a s t e r Than  of D i g i t F i v e .  Under Scheme I .  Classific-  /  Figure  3.5  Example o f the E f f e c t o f Addling Small Hysteresis  Zones a t Quantized D i r e c t i o n  Zone B o r d e r s .  33  I.  Without Hysteresis  D I R E C T I O N INFERRED FROM PRESSURE  Q U A D R A N T  D E S C R I P T O R  P >0 , P >0  P <0 , P >0  6 ^ 1  1  3  1  3  1 < t a n 0 < oo  2  2  4  4  -1 Y -±- ) X P  (6 = t a n  0 <_ t a n  LEGEND:  x  A  _ P P  A.  Y  y  x  _ = Pressure  . in  = Pressure  i n Y-direction  v  p  Y  x  !0 .  ,. X-direction  6 = I n f e r r e d d i r e c t i o n measured horizontal, i.e.  c o u n t e r - c l o c k w i s e from  the  ft Other = P r e v i o u s d i r e c t i o n not i n Quadrants therefore II.  quantize ignoring  d e f i n e d by b o r d e r  hysteresis  With Hysteresis  D I R E C T I O N INFERRED FROM PRESSURE  Q U A D R A N T P >0  ,  x  P (6 = t a n "  1  X  0 <_ t a n .754<tan 1.327  P >0 y  PREVIOUS QUADRANT  - ^ j  D E S C R I P T O R P  X  >  0  '  P  Y -  0  p  1 0 * ±0 P  x  Y  PREVIOUS QUADRANT  PREVIOUS QUADRANT  PREVIOUS QUADRANT  1  2  OTHER  3  2  OTHER  1  4  OTHER  3  4 OTHER  6 <_ . 754  1  1  ft  3  3  *  1  1  ft  3  3  0 ^1.327  1  2  ft  3  2  *  1  4  *  3  4  2  2  *  2  2  ft  4  4  ft  4  4  <_ t a n  6 < °°  *  ft  where t a n 45° - 8° = 0.754 t a n 4 5 ° + 8 ° = 1.327 Figure  3.6  Summary o f Q u a d r a n t Q u a n t i z a t i o n Schemes  P Y  ±0  i n implementing the  hysteresis  Chui suggested o c t a n t s were  feature  i s offset  of unequal s i z e  variations  particular  [CHUI76].  for which  obtained using a plane with or v e r t i c a l strokes  are  1,2,3  the  a n d Down(D)  the in  pen i s the  In order raised  as  defined  to i m p r e c i s e h o r i z o n t a l  calculated  direction-code  referred  t o as R i g h t  Segments  (R) , UP (U) , L e f t . ( L )  d i r e c t i o n codes  are  important.  "RDLDRO".  from the w r i t i n g medium.  photograph,  (b),  character  t a k e n o f when  information is  of recognizer  and so  trace,  In Plate  The f i n a l  included  (c)  to draw " 0 " then  and  The  the  A t the  (a),  the  "0" indicates  a rightward  (R) s t r o k e  present time-ordered that  the  followed  on.  numeric " 3 " , i s e x p l a i n e d i n a s i m i l a r  Plate  output.  Such a sequence i s i n t u i t i v e l y  f o r d i g i t two w h i c h b e g i n s w i t h  Plate  a multi-stroke  d i s p l a y e d on the Megatek CRT.  d i r e c t i o n sequence i s  by a downward s t r o k e  and t h i s  a derived composite pressure  time,  p e n was l i f t e d  describe  appear three photographs  d e c i s i o n are  only the  do n o t  vector.  classification  instructed  Results  r e - l a b e l e d i n a more m e a n i n g f u l f a c h i o n . hereafter  the  4.2.7.  of the  from the w r i t i n g s u r f a c e  direction-code vector,  final  tolerance  to improve symbol d e s c r i p t i o n , note i s  I n F i g . 3.7  reasonable  increased  r e c o g n i t i o n system i s designed.  an i n c r e a s e d  are  codes  direction-code  quantized  accommodate  octal  respectively.  Direction completely.  to  obtained.  i n which  T h i s i d e a c o u l d be t a i l o r e d , t o  w e r e made a w a r e  quadrants were  and 4 o f F i g . 3 . 5  enlarged  examined i n S e c t i o n  Since subjects vector,  the  results  Under h i s implementation, the  of oblique l i n e s .  symbol set  superior  using a non-uniform quantization plane  segments c o v e r i n g d i a g o n a l d i r e c t i o n s were direction  by the  is  a l i t t l e more c o m p l i c a t e d .  to take  the pen from the  paper  fashion.  The  The w r i t e r  was  and t o  cross  a  35  SPECIMEN A SLOPE CODE VECTOR  SLOPE CODE VECTOR ROLftOLO  PRESS.INTERPRETATION OF, CHAR. 0  PLATE  Figure  3.7  Examples of T r a i n i n g Program's  Output  (c)  diagonal l i n e through oersted).  it  The p e n - o f f  character  c a l l e d the  t e c h n i c a l zero or  on two o c c a s i o n s ,  symbol completion; t h i s  explains  the  once b e f o r e  direction example,  codes the  are  scheme.  It  merely v e r i f i e s  i d e n t i c a l and d i s c a r d s  p a r t i a l l y processed  "RRRDDLDDDRO", w o u l d b e r e d u c e d Now t h a t  the  t h a t n o two  one o f t h e m i f t h e y  direction-code vector to  reader i s  perspective.  Plate  are.  for d i g i t  (B) d e p i c t s  rightward stroke testing  are  evident  in Plate  end-points  the  the  aware  o f how d i r e c t i o n c o d e s  detected  is  ( C ) ; the  omitted.  also performed.  the  processor's  "DRULDRULDRDR" i s  The r e s u l t  the  3.4  final  stage of the  The C l a s s i f i c a t i o n  3.4.1  output  l o n g and c o m p l e x . the  final  e l i m i n a t e d . . X - , Y - and  initial  I n the  determined,  different  and f i n a l  fourth example, is  the  recognition process,  Z-threshold  downward  strokes  clipping  d i r e c t i o n code  " L D R U L " , n o m i n a l l y d e s c r i b i n g n u m e r i c s i x and i n a form s u i t a b l e to  two,  are  of applying Z-threshold testing;  i n Specimen A i s  applied i n Plate (B) a r e  illustrates  d i r e c t i o n code,  effect  For  "RDLDRO".  (A) o f F i g . 3 . 8  pre-processing;  multi-  consecutive  d i g i t s i x a n a l y s i s p r e s e n t e d e a r l i e r may b e v i e w e d f r o m a  Plate  the  direction-code  c o d e r e d u c t i o n was m e n t i o n e d w h e n d i s c u s s i n g t h e  level pre-processing  before  the  LDRUOLO". Direction  the  is  condition i s noted  d i a g o n a l and a g a i n a t vector"  (this  at  stroke  vector, for  entry  classification.  Scheme  Introduction  The f i n a l  phase i s  the  classified into  d e c i s i o n - m a k i n g stage i n which the  sample i s  finally  one o f a f i x e d  classes.  A v a r i e t y of c l a s s i f i c a t i o n techniques  number o f  unknown  pre-defined  have been used i n the  past,  ate (B) S P E C I M E N .A SLOPE  CODE  VECTOR  DRULDRULDRDR'  PRESSURE INTERPRETATION OF CHARACTER* 5  SPECIMEN  SLOPE  CODE  C  VECTOR  RULDRULDR  PRESSURE INTERPRETATION OF C H A R A C T E R 5  Figure  3.8  •SPECIMEN  SLOPE  CODE  B  VECTOR  DRULDRULDRD  PRESSURE INTERPRETATION OF C H A R A C T E R 5  SPECIMEN  SLOPE  CODE  D  VECTOR  LDRUL  PRESSURE INTERPRETATION OF CHARACTER E  Examples Demonstrating the E f f e c t o f M u l t i - l e v e l Pre-processing." (Specimen A i s t h e raw sample, specimen D i s the f u l l y pre-processed version.)  38  a number of them are summarized  below.  The table look-up method necessitates a perfect match of input character features to one of a number of pre-defined prototypes.  In general,  c l a s s i f i c a t i o n performance depends d i r e c t l y upon the number of table e n t r i e s . Since the form of handprinted characters i s loosely constrained, table s i z e may have to be p r o h i b i t i v e l y large for the minimal ( i . e . , environment under consideration.  microcomputer)  This statement i s e s p e c i a l l y true f o r a  "many user" environment. The best-match or nearest neighbour approach i s a promising a l t e r n a t i v e since table size may be reduced appreciably.  Although the sample  feature s t r i n g must s t i l l be compared to pre-selected prototypes, the number of comparisons i s reduced since only a subset of a l l variations must be e x p l i c i t l y defined.A "degree of s i m i l a r i t y " test i s applied and the best match (which may i n fact be exact) i s chosen; M i l l e r [MILL69] and I'chikawa  [ICHI74]  apply the Lee distance metric [BATC 74] to this end. However, because of the inherently non-parameterized or unstructured nature of the best-match method, the search for a best match usually requires comparison against a l l prototypes.  Furthermore, the c a l c u l a t i o n of the  defined measure of s i m i l a r i t y i s more complex than a simple exact-match comparison.  These two factors contribute to a slow system response .  The author chose to pursue a d i f f e r e n t approach.  Unknown samples  are c l a s s i f i e d by having successive pen d i r e c t i o n data move a pointerthrough a t r e e - l i k e structure.  As pressure information i s analysed, the pointer  i s moved increasingly closer to the f i n a l c l a s s i f i c a t i o n decision.  At l e a s t  one other author has implemented a simpler hardware version of t h i s scheme and reports moderate success [CRAN77]. given below.  J u s t i f i c a t i o n for such a choice i s  39 In and f a s t match  a real-time  system response  desirable.  a l l character the major  Furthermore, (or  required per  is  concurrently with  stroke  forms.  chronological stroke where as  a final  data  are  the to  After  treated  as  D i c t i o n a r y F o r m a t and  direction is  than  non-trivial begun  Poor system response The p r o p o s e d  f o r an ANSI  is  scheme  numerics  only four  final  characters  c l a s s i f y i n g each  update  pointer are  stroke,  into  their  the  entire  a second stage o f c l a s s i f i c a t i o n  In this  by " p e n - o f f "  thesis,  characters  comprises  a series  indicators  are of  s i m p l e and  treated  inferred  (represented  a form o f d i r e c t i o n code i n f o r m a t i o n .  by  "0").  This  time-efficient.  Operation  As a c h a r a c t e r  is  information is  generated.  for this  character  process.  divide input  to  exact-  n o r m a l l y be  average the  usage  A best-match  repeated  classifier.  250 w o r d s  acquisition  drawn,  a continuous  stream o f i n f e r r e d  E x a c t l y how t h e  r e c o g n i z e d as b e l o n g i n g to one o f the  discussion  available.  and a l l e x c e p t  rendered.  punctuated  cannot  good s i n c e on t h e  form o f u n i - l e v e l c l a s s i f i c a t i o n i s  3.4.2  are  The d i r e c t i o n - c o d e v e c t o r  d i r e c t i o n s w h i c h are Pen-off  usually high.  to perform the  sequence i s passed  decision is  single units.  data  character  Many a u t h o r s c h o o s e associated  are  core  the  l i s t i n g of possible  comparisons  requiring less  System response  updates are performed  stroke)  efficient  As i n d i c a t e d above,  drawback o f a b e s t - m a t c h  very core e f f i c i e n t  dictionary.  environment,  an e x h a u s t i v e  a m a j o r CPU e f f o r t  metric calculations.  therefore  resource  a n d t h u s memory r e q u i r e m e n t s  scheme n e c e s s i t a t e s  is  are  technique necessitates  variations  until  limited  pre-defined  direction-code classes  is  pen sequence  the p o i n t  section.  The d i c t i o n a r y ' s s t r u c t u r e  is  representable  by a m o d i f i e d  tree  of  40  structure  i n w h i c h e a c h n o d e i s s u c c e e d e d b y a maximum o f s i x o t h e r  nodes.  The  are l o g i c a l l y  to  nodes  visualize feature  c o n n e c t e d by d i r e c t i o n  a node p o i n t e r w h i c h ,  extraction stage,  identification  of the  is  lines  codes  represent  are  i s useful  codes are passed  guided through the  tree  until  from  the  a fairly  positive  f o r ANSI n u m e r i c s i s i l l u s t r a t e d i n F i g . 3 . 9 .  the  t o p o f t h e node i n t h e  a r e node c o n n e c t o r s and a r e  direction  It  u n k n o w n s a m p l e may b e o b t a i n e d .  A p o r t i o n of tree Horizontal bars  as d i r e c t i o n  codes.  reported.  followed  linked-list.  o n l y when t h e i r  Numerics i n c i r c l e s  Vertical  associated  represent  the  classification  decision. It  i s important  to r e a l i z e that  a direction-code vector defined  one c h a r a c t e r may b e a s u b s e t  of that defined for another.  subset  v e c t o r i s i n p u t i n the  u s u a l f a s h i o n and t h e p e n i s  raised  from the p a p e r .  outputs  the  direction probable fore,  I f no f u r t h e r  force data are  symbol c o r r e s p o n d i n g to the  information t h a t the  is  subset  subset  character.  c h a r a c t e r was n o t i n t e n d e d .  maximum w a i t i n g t i m e , t e r m e d t h e  the p o i n t e r .  reflecting character  the  user  takes  modified  subset  character  F i g . 3.9,  i s output.  the base of the  classifier  subsequent  In  The d e c i s i o n ,  there-  obtained.  i n order to  the  time-out  NODE B u n t i l  the  further decision  tree diagram the  subset  link. set  to  "START".  t o w r i t e i n a "down" d i r e c t i o n ;  and remains at  is  time-out d e l a y , bounds  the p o i n t e r i s i n i t i a l l y  t h e pen and b e g i n s  accordingly  If  I f data i s not received a c l a s s i f i c a t i o n  i s l o c a t e d at In  the  r e c e i v e d , t h e p o i n t e r moves e l s e w h e r e a n d i t  p e r i o d d u r i n g w h i c h m e a n i n g f u l d a t a m u s t be d e t e c t e d displace  the  subsequently  detected,  i s b a s e d u p o n w h e t h e r o r n o t new d i r e c t i o n e o d e s a r e The  Assume  for  a direction  Suppose  the  the p o i n t e r other  than  is  down  Node A d o w n Node B  9 h Node C p  e n o f f  I  u P  e f t  WNode D  WNode E  d o w n  M  Node F  P e n  9 h  o f  Figure  Node G pen off  3.9  Example  of  Structure.  the  C l a s s i f i c a t i o n  Tree  42 is  reported.  reference  I f the pen i s  an o u t p u t  feedback  circle  from the p a p e r ,  and d i g i t  one i s  the p o i n t e r  transmitted  is  updated  to the w r i t e r  are  other  than  c o n s i d e r e d redundant  "D" w i l l  to  the  not  affect  the p o i n t e r ' s  repertoire.  The f i r s t  subsequent pen-off  right stroke  c o d e t o NODE D .  started.  Sampling continues  inferred,  the  c l a s s i f i c a t i o n process  underlines  since the  another  data  as w e l l  characteristic the w r i t e r i s  are  are  a l l are  top  of the  of the  eliminated.  dictionary; i t  the  considered final  "R"  i s explained next.  intended strokes  are  of the  pre-processing but  The d i c t i o n a r y p r o v i d e s  data  one,  is  structure.  a l l o w i n g the p e n - t i p  as a consequence  is  writer  to  formed, strain-gauge  A large percentage of these u n i n t e n t i o n a l  case of d i g i t  and s p u r i o u s  data  T o l e r a n c e to the  and a f t e r  a  a downward s t r o k e  communicated to  cautioned against  discarded during front-end  and i n t h e  variant  one i s  character  time counter  as the n o m i n a l " D O " , i s  i n v a r i a b l y generated  high sensitivity.  variations  safety  "DRO",  the w r i t i n g medium b e f o r e  spurious  digit  resulting  t o NODE C a n d  a software  and u n l e s s  of h a n d p r i n t e d numeric one.  Although contact  At t h i s point  i s r e p o s i t i o n e d to the  Notice that representitive  updates the p o i n t e r  uninterrupted  system "times-out",  and t h e p o i n t e r  not  for  position;  d i r e c t i o n - c o d e v e c t o r w o u l d n o t b e a v a l i d member o f t h e n o m i n a l  array's  to  purposes.  Directions they  lifted  ;  occasionally  a second degree of  "DRO" i s a c c e p t e d  " R " i s deemed i r r e l e v a n t  as  an a l l o w e d  to the c l a s s i f i c a t i o n  process. A similar data  "RU" are  structure  is  e x a m p l e a p p e a r s as P l a t e  generated unaffected  at  (B) i n F i g . 3 . 1 0 where  the base of the handprinted  and a c o r r e c t  "9".  recognition decision is  spurious  The d e c i s i o n reported.  SLOPE? CODE VECTOR} UROLROLO  OF CHARACTER J  PLATE  PLATE  Figure  (B)  3.10  (A)  Intentional  Stroke  U n i t e n t i o n a l Spurious  Examples o f S t r o k e  Formation  Data  End-Point  (URDLRDLO)  (LDRUDRUO)  Data  Even i n t e n t i o n a l strokes A nominal d i g i t Fig.  t h r e e i s begun w i t h  3.10 depicts  variations  are  performance  that  tolerated  order  a l l input  to  characters forming  represent  irrelevant.  at  is  not  (A) o f penalized;  the c l a s s i f i c a t i o n set.  system response  attempts  i s understood,  when w o r k i n g w i t h  as  time i t  is  assumed  forming characters  of  the  f o r example,  that w r i t e r s w i l l  a numerics only d i c t i o n a r y . the  not  Consequently,  user has  finished  them; enough i n f o r m a t i o n had a l r e a d y been r e c e i v e d t o e x c l u d e of a l l other  truncated  structure  s y m b o l - s e t members.  character five  indicated.  p o i n t e r moves t h r o u g h character,  the  user  This idea is i l l u s t r a t e d in  d i c t i o n a r y ' s inherent  forms appears i n F i g . 3 . 1 2 . is  the  of F i g . 3.11.  An example o f t h e  digit  user  defined character  decrease  honest  It  N o t i c e how t h e  c a n o f t e n be c o r r e c t l y r e c o g n i z e d b e f o r e  possibility the  further  classified  they begin to i n f l u e n c e  members o f t h e  defined symbol-set. enter letters  until  sometimes  a rightward stroke but P l a t e  an upward s t r o k e .  of other  In  are  the  lifts  tolerance  A decision tree  As the w r i t e r forms the  first  t r e e from "START" to n o d e . E . t h e p e n a n d makes t h e  for  to non-nominal the  stroke,  2-stroke "b"  }  To c o m p l e t e  necessary  the the  rightgoing  stroke. Upon o b s e r v i n g r a w d a t a , experienced user  could  form  the  it  became e v i d e n t  second s t r o k e ,  that only  the  a simple "RO", c o r r e c t l y .  M o s t u s e r s u n i n t e n t i o n a l l y p r o d u c e d some f o r m o f s p u r i o u s e n d - p o i n t Direction-code vectors The Note  recognition logic that  pointer  such as following  "URUO",  "URDO",  node E a c c o m o d a t e s  i f a particular direction link  i s not  and  does  not  u p d a t e d when s u c h a d i r e c t i o n i s  data.  " L U R O " w e r e common. a l l such issue  combinations.  from a node,  registered.  the  45  Start  '0 'D 0 Example  0'  R  (A)  F u l l Recognition Logic  "0  R 0  R' '0 Example  (B)  Truncated R e c o g n i t i o n L o g i c . A C l a s s i f i c a t i o n Decision i s Rendered Before the Symbol i s Completed by the User.  F i g u r e 3.11  An I l l u s t r a t i o n o f Inherent D i c t i o n a r y Compactness. (Logic f o r S p e c i a l C h a r a c t e r  ,  RODORORO.)  Node A  W D W Node  B R W Node  C D W  L Node  E  P Node F R  VNode G  F i g u r e 3.12 A Demonstration o f the D i c t i o n a r y ' s Tolerance t o Symbol V a r i a t i o n s (Numeric,"5").  Legend: D=down l;= left P- pen-off H'= right  Node P  By not having to s p e c i f i c a l l y provide for a l l p o s s i b l e v a r i a t i o n s the dictionary structure tends to be compact and readable.  Since  valuable  time i s not wasted updating the node pointer in the case of i r r e l e v a n t data system response time i s very  low.  Once the system designer has decided upon a symbol-set and  any  variations that should be tolerated, a tree i s drawn and the l i n k e d - l i s t constructed.  The design of a standard numerics dictionary requires  mately 5 man-hours.  approxi  Modifications can usually be effected i n a matter of  minutes. Data structure formatting of the l i n k e d - l i s t i s f a i r l y a r b i t r a r y . Before some experience with tree construction had been gained, a 16-bit word per node entry was  utilized.  to anywhere within the l i s t . l i m i t e d jumps to  This allowed pointer updates from and  Later, an 8-bit per entry format which  63 entries was  adopted.  Even extensive  d i c t i o n a r i e s can  be constructed under t h i s constraint. There are s i x sequential entries per node and nodes may entered i n any order.  3.5  Two  be  of the formats used are i l l u s t r a t e d i n F i g .  3.13  Testing at Intermediate Stages  3.5.1  Spurious Data at Stroke End-Points  In Section 3.3.2, i t was  indicated that pen-tip p o s i t i o n i s not  r e a d i l y extracted from force information.  Precise character forms cannot  be reconstructed yet approximate shapes adequate f o r diagnostic purposes may  be derived with the use of a graphics display system. I f pen-tip pressure  i s .assumed d i r e c t l y proportional to cursor  v e l o c i t y , processed force data may dinate increments.  This data may  be applied to determine d i s p l a y  coor-  then be used to drive a displayed trace  48  (A)  List  Structure  n o d e ( i -1}  node  (i)  node {  P r e f e r r e d Format f o r 8-bit, Microcomputer  (B)  - 1 Byte per L i s t  entry (k)  Entry  entry (k+1)  S i g n - B i t f o r +63 ?able Jump F l a g (C)  Table Jump  A l t e r n a t i v e Format ; One 1 6 - b i t Word Per L i s t E n t r y  e n t r y (k) e n t r y (k+1) sign/type  flag  00 01 10  Figure  3.13  meaning P o s i t i v e o r Zero Jump Negative Jump A s c i i Flag  Linked-List  Entry  Formatting  49 cursor.  F i g . 3.7  end-point data  data  illustrates  are w e l l  s e v e r a l c a l c u l a t e d symbol t r a c e s .  represented;  their  s l o w e r movements  Spurious  g e n e r a t e many  points. With  identify  a i d of the  and c o r r e c t  pen posture  3.5.2  the  d e s c r i b e d d i s p l a y system u s e r s were  chronic d i f f i c u l t i e s which often  and vague s t r o k e  stemmed  able  to  from poor  definition.  D i r e c t i o n - C o d e V e c t o r C a l c u l a t i o n f o r System  ' a r a m e t e r and D i c t i o n a r y  entry Optimization Before extraction  techniques  were i n v o l v e d at with  this  s p e c i a l emphasis  intermediate in  d e s i g n i n g the  recognition logic,  were t e s t e d  rigourously.  l e v e l of testing. p l a c e d on the  t e s t i n g was f r u i t f u l  pre-processing  A p p r o x i m a t e l y 15  Various character  d i g i t s zero  (Section  d i c t i o n a r y used i n the  were  requested  to t r y  to  final  stage o f  improve t h e i r  currently,  the  for  reference.  future  generated  recognition scores;  were most  later  the  tabulated  A subsequent examination i n f l u e n c e d the and h e l p e d d e f i n e w h i c h  character  of  Subjects character Con-  and  stored,  final  select-  variations  common. In  users'  construction  for well-drawn forms.  d i r e c t i o n - c o d e v e c t o r s were  ion of dictionary prototypes  T h i s form o f  ranges used  the proposed system.  trace provided strong p o s i t i v e reinforcement  selected  4.2).  The d i a g n o s t i c p r o g r a m was a l s o i n v o l v e d i n t h e the  feature  subjects  s e t s were  through n i n e .  i n i d e n t i f y i n g parameter  d i c t i o n a r y p e r s o n a l i z a t i o n experiments  and  the n e x t  performance  are  section several described.  other  techniques  applied to  improve  50 3.5.3  Development o f U s e r - T r a i n i n g A i d s  Since non-laboratory  a d i s p l a y processor would probably  i m p l e m e n t a t i o n a n a l t e r n a t e d i a g n o s t i c t o o l was  F o u r L E D i n d i c a t o r s w e r e m o u n t e d on t h e h a r d w a r e configured  t o m o n i t o r and d i s p l a y i n s t a n t a n e o u s  A photograph in  the  uniform quadrant  this  strain-gauges  t h e pen was moving t o the users,  from the  to the w r i t i n g medium.  condition.  would i n f e r  resulted  direction signals  right.  the  nominally v e r t i c a l attitude o f the  frames  i s shown, i n A p p e n d i x C .  user  forgetting  to  keep  The i n d i c a t o r s w e r e the  left  t o t h e LED a r r a y  alleviate  w e r e made a v a i l a b l e t o  legs  paper.  Without e x c e p t i o n system performance  frame d e f i n e v e r t i c e s  t h e new d e v i c e .  The s u p p o r t  u s e r h a d g a i n e d some e x p e r i e n c e  to p r i o r claims of a natural Subjects  were  asked  when c o r r e c t l y s u p p o r t e d .  and  useful  the  even, i f for  some  eventually  summarized i n Chapter V . the  defined  the  of a t r i a n g l e  users.  and i n  The  a  pen-tip  in, the plane  of  the  i m p r o v e d s i g n i f i c a n t l y upon  was r e m o v e d , h o w e v e r ,  in writing  since i t s  as soon  use was  as  contrary  device. to  take  Suggestions  and r e s h a p i u g were s o l i c i t e d .  evaluating  arranged  t h a t h e l d the pen f i r m l y  and the  the  direction.  difficulty.  Two-point metal support  introducing  and  S i n c e t h i s p r o b l e m was r e c u r r e n t  a s p e c i a l t r a i n i n g - a i d was d e v e l o p e d t o  eliminate  module  W i t h the pen l e a n i n g towards  left  a  devised.  showing the L E D ' s  configuration described e a r l i e r  the pen t i p p e r p e n d i c u l a r detecting  interface  quantized pen  o f t h e pen and s u p p o r t i n g hardware  Poor r e c o g n i t i o n often  in  n o t be a v a i l a b l e i n  fitted  regarding pen-barrel  Comments r e c e i v e d w e r e n o t e d  S i n c e at system,  n o t e o f how t h e p e n  this point  the  hand  repositioning and  are  the main concern  was  n o a t t e m p t w a s made t o m o d i f y p e n  hardware.  51 IV.  4.1  F I N A L EVALUATION OF THE COMPLETED RECOGNITION SYSTEM  Classification with  Experiments were performed  a Restricted Dictionary  r e l a t i n g to  on t h e  trials, those  investigation. to  define  factors  o b t a i n e d were  not  statistical  It  under  regarded  considered.  other  of general  character  appears as F i g . 4 . 1 .  and more t h a n  set  statistically  which  results to  c o r r e c t method  of  set.  for handprinting  tabulated  i n Chapter  be  Institute  f a c i l i t a t e d comparison w i t h sheet  i s i d e n t i f i e d and c a t e g o r i z e d Results of  I,  had t o  An e x a m p l e o f a c o m p l e t e d s c o r i n g  Each subject  four  sessions,  3n (greater than — ) set o  of  as  each  comprising  handprinted  recorded. of  seven experiments  four thousand  number w o u l d have  in  the American N a t i o n a l Standards  3 t h a n —") a n d a l a r g e o  A total  Any  a  repetitive  r e c o g n i t i o n scheme p a r a m e t e r s t h a t  b e l o n g i n g t o one o f t h r e e g r o u p s .  A N S I . d i g i t s , were  i n an e n v i r o n m e n t  system c h a r a c t e r i s t i c s  c l a s s i f i c a t i o n schemes.  a small (less  test  to perform  c o u l d be h e l d c o n s t a n t .  to each d a t a  Experimenting with  (ANSI) p r o p o s e d  others necessitated  S p e c i a l c a r e was t a k e n t o a p p l y the  analysis  a number  and t o  I  objectives  as s t a t i s t i c a l i n n a t u r e a n d a s s u c h s u b j e c t  A s i d e from the t h e r e are  evaluation,  was c o n s i d e r e d i m p o r t a n t  analysis  outlined i n Table  A l t h o u g h a number o f  for their  c o n t r o l groups,  misinterpretation.  design objectives  completed system.  r e q u i r e d l i t t l e o r no t e s t i n g thorough  the  rendered  were  c o n d u c t e d on t w e n t y i n d i v i d u a l s  d i g i t s were p r o c e s s e d . results  less  i n s i g n i f i c a n t ) because o f  meaningful  A reduction i n (and p e r h a p s  the h i g h sample v a r i a n c e  sample  even encountered.  GENERAL  DICTIONARY*  CONSTRAINTS System  PLACED  Test  SUBJECT CATEGORY  Professional/ Technical  TESTING UPON THE  USER | W r i t i n g  Parameters  SU3JECT  Subject  VI  hone  Medium:  none  Pressure  Threshold:  6  Time-cut  value  Z  Pressure  Threshold:  6  Clipping  Time  c  E S S  SIZE  SMALL  -  TEST DIGIT  rw  zero  H  X X X  six  X  seven  X X X  Sub-Total £  6  .  4  0  U  C X  X X X  X  X  X  X  X  X  X  X  X  X  X  X  X  X  X  X  X  9  standards  1  0  10  X  X  X  X  X  X  X  X  X  X X  X  X  X  0  X X  x  0  X  X  X  X  U  M  X  X X  X  X  X  six  X  X  seven  X  X  X  X  eight  X  X  X  X  ni ne  X  X  X  X  Sub-Totals  8  2  0  9  1  0  8  2  0  9  1  0  TOTALS  14  6  0  17  3  0  17  3  0  19  1  0  Legend: t o ANSI  C  X  0  0"  4 U  X  five  F i g u r e 4.1  M  X  2  f o r handprinted  characters  Example o f a Completed R e c o g n i t i o n  as S  B E R  X  8  b)zero  (deg's.):  3  one  four  conforming  Non-linearization  zero  three  *  conforming  X  two  LARGE S SHALL  M  a)ANSI  Plane  N U M  X  eight n ine  1 0 N  X  X  five  LARGE  C  X  four  (sec.):0.08  X  three  i  U  X  two  Stylistic: (sec.):0.6  2  1  one  1  Writing  X,Y-  CHARACTER  IDENTIFIER  Speed:  C -  classified  M * U •  rnisclassificd unknown t o t h e  correctly dictionary  Performance S c o r i n g Sheet  53 ,  The s e v e n q u e s t i o n s i)  dictate  iii) iv)  Is  the  operating the  user-category  differences  Is  training is  is  the pages  character-size  required before  variations? over  the  generate the  performance  o f 95%?  are m i s c l a s s i f i c a t i o n s  to certain characters?  acceptable  that  inter-  significant?  accepted l e v e l  "trained",  response,  How o f t e n d o e s  training or is  follow,  an  "UNKNOWN"?  system performance contingent  continual user  In  are  the  user?  the w i d e l y  Once a u s e r  input  i.e.,  to  r i g h t - h a n d e d u s e r have an advantage  How much u s e r  limited  vii)  system?  performance s e n s i t i v e to  Does t h e  system  c l a s s i c a l e x p o n e n t i a l l e a r n i n g curve?  type of user  Is  are:  users  Does t h e  r e c o g n i t i o n performance d i r e c t l y r e l a t e d  attains vi)  f o r most  i n some p r e d i c t a b l e m a n n e r ?  left-handed v)  a n s w e r s w e r e sought:  Does s y s t e m r e c o g n i t i o n p e r f o r m a n c e progress  ii)  for which  the  skill  upon  easily  e x p e r i m e n t s p e r f o r m e d and r e s u l t s  retained?  obtained  are  described i n d e t a i l .  4.1.1  The L e a r n i n g  The with  the  groups  first  Curve  section of subject  familiarization  f o r the purpose  upon u s e r  type  process.  performance  evaluation i s  Subjects were d i v i d e d i n t o  concerned  three  of i n v e s t i g a t i n g r e c o g n i t i o n performance  specific  dependence  ( i . e . , p r o f e s s i o n a l / t e c h n i c a l , p r o f e s s i o n a l / n o n - t e c h n i c a l , and  non-professional/non-technical).  Throughout t e s t i n g subjects  were  asked  to  a d h e r e t o ANSI c h a r a c t e r  forms w i t h  be w r i t t e n as an o e r s t e d  (0).  Of of  the  particular  subject  c o u l d be k e p t as  technical  this  is  categories. fairly  brief  exception of d i g i t  z e r o w h i c h was  was t h e l e a r n i n g c u r v e f o l l o w e d b y  I t was found t h a t  the  familiarization  and t h a t n o n - t e c h n i c a l  users  to  each  period  f u n c t i o n e d as  well  users.  The category  interest  the  average  r e c o g n i t i o n performance  shown i n T a b l e I I I .  F i g . 4.2 i s  f o r each t r i a l  i n each  subject  a graphical representation,  of  data.  TABLE  III  AVERAGE RECOGNITION PERFORMANCE  SUBJECT/  AVERAGE RECOGNITION PERFORMANCE (%)  CATEGORY  TRIAL  Prof./ Tech.  1  TRIAL 2  TRIAL  3  TRIAL  4  56.0  73.1  79.1  88.1  Prof./ Non-Tech.  39.2  64.2  75.8  90.0  Non-Prof./ Non-Tech.  43.8  60.0  76.3  87.5  WEIGHTED AVERAGE  50.2  69.1  77.9  88.5  1 0 M I N . OF TRAINING  15 M I N . OF TRAINING  NO TRAINING  It be e x p e c t e d  is  clear  5 M I N . OF TRAINING  from the poor r e s u l t s  to perform adequately  of t r i a l  w i t h o u t some u s e r  1 that  the  system  cannot  training.  ' n o n - t e c h n i c a l ' i s t o mean t h o s e s u b j e c t s who do n o t p o s s e s s k n o w l e d g e o f how t e c h n i c a l d e v i c e s o p e r a t e  detailed  A L e a r n i n g Curve i s Merely the Smooth F u n c t i o n Drawn Through the 20  P o i n t s Representing Successive  Average R e c o g n i t i o n Performance f o r Each  100  Trial  18 -  90  16 -•  ..80 Technical  -70  X  12}-  Non-Technical (A Single Curve Represents Data Drawn from Performance Scores of Both Professional and Non-Professional Subjects)  -60  10 ••  •50 \  Professional/Tech.  O X  •L0  Professional/Non-Tech. •30  Non-Prof./Non-Tech,  •20 ••10 Trial 1  Trial 2 5  F i g u r e 4.2  Trial U  Trial 3 10 lotal time spent in training  (mins.)  A L e a r n i n g Curve o f S i n g l e T r i a l R e c o g n i t i o n Performance  15  As e x p e c t e d , the  average  Improvement  15 m i n u t e s  requested  [CHUI76]. required  Section 4.1.5  the  of t h i s  i n the  final  average  "no-user  appear met,  subject's  advantage,  to suffer  4.1.2  than  after  infer  l e a r n i n g a new  were a v e r a g i n g  skill.  slightly  character  recognition  to d e t e r n u n i n g  is  scheme  the  time  level.  for  15 m i n u t e s  achieves  (56.0 versus  each o f t h e  best:  3 9 . 2 and  other  of t r a i n i n g i s not  (88.1 versus  the  43.8%),  categories greater  a l t h o u g h c e r t a i n u s e r s may h a v e after  a brief  an  training does  A s y s t e m d e s i g n g o a l seems t o h a v e reflects  the n a t u r a l  A more r i g o r o u s p e r f o r m a n c e  than  88.5).  the n o n - p r o f e s s i o n a l / n o n - t e c h n i c a l s u b j e c t  any h a n d i c a p s .  the  a 95% m i n i m u m p e r f o r m a n c e  i s devoted  that  that  initial  conform e x a c t l y to  a l l users perform equally w e l l  environment/interface  technical user. next  thesis  performance  In p a r t i c u l a r ,  the  literature  training" session  less  performance  Sample r e s u l t s  period.  however,  practice,  88% o v e r t h e  the p r o f e s s i o n a l / t e c h n i c a l subject  t o t a l improvement i s  initial  I n the  to achieve t h i s performance  performance  and h i s  does n o t ,  improves w i t h  from 50 to  for a real-time handprinted  Although  his  rose  of t r a i n i n g , subjects  t h a n 90% r e c o g n i t i o n .  usually  performance  e x p o n e n t i a l f u n c t i o n s i n c e users are not After  less  subject's  r e c o g n i t i o n performance  15-minute p e r i o d . classical  the  tendencies  analysis  been  of the  i s undertaken  not  non-  in  the  section.  Performance S e n s i t i v i t y to User Category  The p u r p o s e performance  for users  of this  section is  from each c a t e g o r y  t o c o m p a r e r e c o g n i t i o n scheme ; once they  are  familiar  with  the  57 system's  operation.  In particular,  15 m i n u t e s  of t e s t i n g  analyzed.  Recognition scores  the hypothesis  i s independent  of subject  f o r each  o f the  that performance  c a t e g o r y w i l l be three categories  after  statistically appear  in  Table IV. TABLE I V PERFORMANCE A F T E R 15 MINUTES OF T R A I N I N G  S U B J E C T PERFORMANCE FOLLOWING 15 MINUTES OF T R A I N I N G (%)  CATEGORY MEAN PERFORMANCE  Professional/Technical  95 95 95 95 75 85 80 90 95 85 90 95 80 85 80 90  88.1  Professional/Non-Technical  90 85 90 95 85 90  88.3  Non-Professional/ Non-Technical  85 90 85 90  87.5  CATEGORY/ SUBJECT  Since obtained performance from  is  subject  to  category  means  accepting  the hypothesis  performance  implicitly  i n f l u e n c e d b y human  fluctuations.  F o r example,  for professional/technical  starting  differ.  Unfortunately  are  chance  75% t o 95% w e r e r e c o r d e d A reasonable  of  data  point  is  that  the  performance  variance  variation within  the  scores  is  independent  o f s a m p l e means category  2  (s_)  and w i t h o u t  ranging  subjects.  a c a l c u l a t i o n of the  A s m a l l variance would strengthen  judgement,  the of  degree  to  claim  for  which  category.  indicates further  nothing  knowledge o f  2 subject  performance  fluctuations  (i.e.,  w o u l d be u n w i s e t o make a n y i n f e r e n c e s parameters.  subject  performance  variance,  regarding underlying population  )  it  The n u l l h y p o t h e s i s must b e i n t r o d u c e d a t random v a r i a b l e t h a t d e s c r i b e s variables. category  Let  A  scores.  hypothesis  A  d i f f e r e n c e between  > in particular,  x C  the  T  The c o n d i t i o n Ax_  that performance  i s not  this  represent  = 0  the  time.  L e t A be  two o t h e r  random  difference  i n mean  denotes the n u l l h y p o t h e s i s ,  related  to s u b j e c t  a  category,  i.e.,  the  that  s a m p l e s were drawn from a s i n g l e s t a t i s t i c a l p o p u l a t i o n . 2 By d e f i n i n g t h e the  average  of subject  independence are  resolves  different,  i.e.,  "pooled subject  performance  variance",  (s  P  )•,  as  2 v a r i a n c e s (s ) , the question of category 2 2 to a comparison of and s^ . I f the p o p u l a t i o n s performance  i f the n u l l h y p o t h e s i s  w o u l d be s i g n i f i c a n t l y  larger  than  S  p  i s not  acceptable,  then  2 s_  point  needs  2  •  Before performing the n u m e r i c a l c a l c u l a t i o n s a f i n a l 2 t o be r a i s e d .  According to  the C e n t r a l L i m i t Theorem  s_  2 s m a l l e r t h a n s^ b y a f a c t o r o f n , t h e number o f s a m p l e s c a t e g o r y [HOEL62]. T h e r e f o r e , i f the r a t i o ,  is  expected  obtained  from  to  be  each  2 ns_ F = 2 X  S  (4)  P  commonly t e r m e d t h e  variance ratio  the n u l l hypothesis  c o u l d be s a f e l y  The v a r i a n c e  ratio's  degrees o f freedom o f F , denominator where of  subjects  r represents  w i t h i n each  f o r m u l a e may b e  (r-1)  categories  the  less  than  1,  accepted.  f o r the numerator  category.  Rigorous proofs  Analysis [WONN72].  dependent  and r ( n - l )  number o f c a t e g o r i e s  f o u n d i n many f o r m a l s t a t i s t i c s  of unequal s i z e  significantly  probability function i s  Table' V is a standard for  [WONN72] i s  upon  in  the  and n ,  the  for degrees of  texts  the  number freedom  [HOEL62].  o f V a r i a n c e (ANOVA)  table  modified  TABLE V  ANOVA TABLE FOR USERS FROM DIFFERENT CATEGORIES  DEGREES OF FREEDOM -  = 2  Z n. (x.-x) 1=1  = 0.1102, i = 1  r - 1 = 2  MSS  1  r  = SS  v  / ( r - 1) K  '  = 0.397  + 0.5116, 1 = 2  MSS  + 0.1731, i = 3 = 0.7949  i I I (x i=l j=l r  n  ± i  _ 2 - x )  =  F =  = SS  27.2 , 1 = 1  Z  1=1  1  (n - 1) = 15 1  +  2.0 , i = 2  +  +  1.0 , i = 3  + __3  =  30.2  = SS  u  5  = 23  MSS  u  = SS / £ (n. - 1) u . , i i=l = 1.313  MSS u  0.3024  ACCEPT THE HYPOTHESIS  This  technique  implicitly  assumes  t h a t samples  are  drawn  from  2 normal populations with may n o t b e t h e  same.  the  same v a r i a n c e , a  Indeed,  it  is  b e i n g i n v e s t i g a t e d t h r o u g h ANOVA The r a t i o , difference  Expressing these results r e c o g n i t i o n performance user  type i s  4.1.3  to 0.3024  and t h e r e f o r e  i n a more f o r m a l m a n n e r , after  the  A number o f s u b j e c t s and t h a t  that  are  at  the  classical  observed  fluctuations.  the h y p o t h e s i s  15 m i n u t e s o f t r a i n i n g i s n o t  of Character Size  drawn t h a n s m a l l ones  their possible differences  i s e a s i l y e x p l a i n e d by chance  c l e a r l y acceptable  Effect  The p o p u l a t i o n means may o r  testing.  F, evaluates  i n c a t e g o r y means  .  that  dependent  5% s i g n i f i c a n c e  upon  level.  on P e r f o r m a n c e ,  felt  that larger  characters  w e r e more e a s i l y  r e c o g n i t i o n performance w o u l d ,  as  a  result,  3 be s u p e r i o r . more o f t e n  The h y p o t h e s i s  that  large  digits  (drawn g r e a t e r  c o r r e c t l y r e c o g n i z e d t h a n s m a l l e r d i g i t s was  t h a n —")  tested.  D u r i n g l e a r n i n g c u r v e i n v e s t i g a t i o n an e q u a l n u m b e r o f and s m a l l d i g i t s were investigated means  through a comparison of large  for this  subject  drawn by each s u b j e c t .  data.  influence  is  desirable  the proposed h y p o t h e s i s  large  The a b o v e h y p o t h e s i s  and s m a l l d i g i t  since subject  is  performance  Such a method o f p a i r e d s a m p l i n g ( i . e . ,  i n both t r i a l s )  variations  u s i n g the do  test.  a l a c k o f i n f o r m a t i o n r e g a r d i n g the p o p u l a t i o n standard  The h y p o t h e s i s character  size,  i.e.,  the  same  not  Because of sample s i z e d e f i c i e n c y (a p r a c t i c a l c o n s i d e r a t i o n )  Gaussian d i s t r i b u t i o n i s  are  r e p l a c e d by the  student's  that performance n u l l hypothesis,  t  deviation,  the  distribution  [WONN72].  i s n o t i n a n y way r e l a t e d is  represented  by H .  and  to  Letting  A represent large  the  and s m a l l d i g i t s ,  Since results alternative the  random v a r i a b l e d e s c r i b i n g the  t h e A=0 c o n d i t i o n r e s t a t e s  i n d i c a t e s u p e r i o r performance  hypothesis, -  Statistically, the  observed  particular,  .  .  the  results  it will  sessions  hypothesis.  d i g i t s , the  one-sided  i s preferable  to  •  would a r i s e  to whether  or not  it. i s  i f the n u l l h y p o t h e s i s were  assuming  were  true,  confidence  a r e p e r f o r m e d on a v e r a g e  exceeds  the  likely  true.  In  the  generally  as r e l a t e d  relation A  recognition scores  1 t o 3 a n d a g a i n on o b s e r v e d p e r f o r m a n c e  to uncover p o t e n t i a l e f f e c t s  t i o n performance  >  to d i g i t  size.  of user  during  t r a i n i n g on  The c o r r e s p o n d i n g  during session classifica-  one-sided  is, D  -  t •°  ^ ^  5  (5)  where  ' D = average t  ^  difference  i n s a m p l e means  = 5Z p r o b a b i l i t y l e v e l distribution  Sp = s t a n d a r d  for,the  Student's  on n - 1 d e g r e e s o f  d e v i a t i o n of sample  n = number o f s a m p l e s  performance  f o r each  category.  hypothesis which indicates  freedom  obtained  E q u a t i o n (5)  o b t a i n e d and a p p l i e d t o c l a s s i c a l  i n d i c a t e d i n Table V I I .  t  differences  T a b l e VI summarizes c a l c u l a t i o n s o f l a r g e  as  for  s i g n i f i c a n c e v a l u e o f 5%.  4 i n an a t t e m p t  data  null  be d e t e r m i n e d i f the p r o b a b i l i t y o f o b t a i n i n g  Calculations training  for larger  q u e s t i o n reduces  observed sample d i f f e r e n c e s , accepted  the  i n performance  , p o s t u l a t i n g a A exceeding zero,  A^O h y p o t h e s i s .  that  difference  R e s u l t s are  is  and s m a l l  evaluated  accept/reject  digit  on t h e b a s i s  hypothesis  testing  statistically conclusive:  t h a t system performance  is  independent  of  the of  null  62  character  size,  for  a l l subject  the  u s e r has  category  is  rejected  during  categories.  completed  the  dependence i s n o t  experimentation  the  i n i t i a l period  However, initial  the  n u l l hypothesis i s  15 m i n u t e t r a i n i n g  implied further  reported i n Section  of system  period.  substantiating  the  familiarization  accepted Once  after  again,  results  of  4.1.2.  TABLE V I MEAN AND VARIANCE OF D I F F E R E N C E DATA  SUBJECT/  AVERAGE OF SESSIONS  CATEGORY  1,2,3  SESSION  4  D = AVERAGE s'A = VARIANCE OF D = AVERAGE s2=VARIANCE OF D I F F E R E N C E (%) THE D I F F E R E N C E D I F F E R E N C E (%) THE D I F F E R N C E #  Professional/ Technical  6.87  1.43  2.50  0.87  Professional/ Non-Technical  7.20  0.73  0.00  2.00  Non-Professional/ Non-Technical  6.60  0  0.00  0.66  DIFFERENCE = S m a l l D i g i t P e r f o r m a n c e S u b t r a c t e d Large D i g i t Performance  from  TABLE V I I C L A S S I C A L HYPOTHESIS  SUBJECT/  AVERAGE OF SESSIONS  CATEGORY  CONFIDENCE RELATION  Professional/ Technical  1,2,3  t DECISION REGARDING  REJECT  A >  -1.16  ACCEPT  REJECT  A >  -0.96  ACCEPT  Professional/ Non-Technical  A >  0.02  Non-Professional Non-Technical  A >  0.66  Accept  or Reject  , CONFIDENCE DECISION RELATION REGARDING Hn ACCEPT  REJECT  ^  (5) at  the  4  -1.59  0.16  to Equation  HQ  SEE5SI0N  A >  A >  Refer  TESTING  c l a s s i c a l 5% s i g n i f i c a n c e  level  63  4.1.4  Right -  and L e f t  -  Handedness  Pen-tip posture performance. from t h a t  a vital  Left-handed persons  role i n successful c l a s s i f i c a t i o n  often hold  assumed by r i g h t - h a n d e d p e r s o n s .  handed s u b j e c t s the  plays  experienced d i f f i c u l t i e s  normal w r i t i n g  technique  f o r the  t h e i r pen at  an. a n g l e  During t e s t i n g ,  with  different  several  left-  t h e p e n and s u g g e s t e d  Xebec pen i m p l i c i t l y  favored  that  right-  handedness . An e x a m i n a t i o n o f r e c o g n i t i o n s c o r e s of both types  o f u s e r s was p e r f o r m e d i n o r d e r  or  their claim.  contradict  professional/technical results  since i t  has  A l l left-handed  category;  this  does  drawn from i n d e p e n d e n t to  statistically  participants not  reduce  already been demonstrated  fell  the  samples  substantiate into  the  generality  that performance  is  of  category  independent. Since population variances As i n p r e v i o u s an u n b i a s e d  they are  estimator,  from b o t h l e f t freedom,  tests  is  are  unknown t h e y m u s t b e  assumed e q u a l .  simply  and r i g h t - h a n d e d  d e f i n e d as groups  estimated.  The p o o l e d s a m p l e the  sum o f s q u a r e d  d i v i d e d by the  total  variance, deviations  degrees o f  i.e., n 2  s P  =  ~ (  n  L  +  1 n  R "  2  )  n„  T  [ £ i-1  (x . ^ T  L  l  - 2Z  x_)  +  Z  (x  i=l ^  - 2  - v ) ] R  (6)  where s n  ^ = p o o l e d sample  P  ,n  ii  =  R  number o f s u b j e c t s  X^^jX^^ = and  ^  ,  X  R  variance  i t h subject  =  a v e r  age  i n groups l e f t  i n groups  left  and  and  right,  right,  performance i n groups l e f t  and  right.  2 For student's  t  the s u b j e c t s  s a m p l e d , s^  distribution applies  f o r the  p r e v i o u s s e c t i o n and i s employed i n the for  the d i f f e r e n c e i n average  was c a l c u l a t e d to be same r e a s o n s following  1.65.  provided i n  The the  95% c o n f i d e n c e i n t e r v a l  r e c o g n i t i o n performance:  where u  =  p o p u l a t i o n mean p e r f o r m a n c e  Xj^,  =  g r o u p mean  t  =  5% c r i t i c a l  li  , u  K  parameters,  performance v a l u e o n (n  - l)+(n  - 1)  degrees o f  freedom s and  n  P  ,n  Since  =  derived pooled standard  =  number o f s u b j e c t s  the  deviation  i n groups l e f t  d i f f e r e n c e between l e f t - h a n d e d  was l e s s  than  2%, t h e  Instead,  the h y p o t h e s i s , H ^ , t h a t  and  right.  and r i g h t - h a n d e d  o n e - s i d e d a l t e r n a t i v e h y p o t h e s i s was n o t -  4 0 , was a n a l y s e d f o r  scores  examined. statistical,  significance. S u b s t i t u t i n g sample d a t a i n t o E q u a t i o n (7) y  L  - y  R  = 0 . 4 ± 1.5  yields,  (  g  )  65 A c o n f i d e n c e i n t e r v a l may b e r e g a r d e d therefore rejected  H ^ , the at  the  alternative  classical  to  =  Training  the s e t  (the n u l l hypothesis)  the u s e r  short  acceptable of  the  o f 90%.  t h i s l e v e l was o f  decision,  i n w h i c h u s e r s are the  the  the  acceptable  usually considered i m m e d i a t e l y made  aware  achieve  o n i m p r o v e m e n t s became  c a t e g o r y was a l l o w e d  l e v e l was r e a c h e d .  users quickly less  P r o g r e s s was  and l e s s  to  initially  grew accustomed to pen o p e r a t i o n .  As t i m e  s i g n i f i c a n t and e r r o r s w e r e  i n g r a i n e d u s e r h a b i t s w h i c h v i o l a t e d the  stroke  traced  formation constraints  of  restricted dictionary. Performance  testing is  reported  f o r each subject  i n Fig. 4.3.  a t t a i n t h e minimum 95% p e r c e n t  statistical  analysis.  specified  level  t h a n one h o u r . h i g h e r l e v e l s as constraints  It  after  5,  1 0 , 15 a n d 3 0 m i n u t e s  I n a d d i t i o n , the  classification  S a m p l e s o f one s u b j e c t  the  fell  interest.  encouraging as the  to  does  left-handed.  amount o f t r a i n i n g r e q u i r e d t o  A s i n g l e v o l u n t e e r from each s u b j e c t  went  clearly  15 m i n u t e s o f t r a i n i n g  S i n c e a minimum 95% l e v e l i s  f o r r e a l - t i m e schemes  until  is  f o r 95% R e c o g n i t i o n  classification  practice  hypotheses,  System performance  is r i g h t - or  Average r e c o g n i t i o n performance a f t e r slightly  of acceptable  5% s i g n i f i c a n c e l e v e l .  n o t e x h i b i t s e n s i t i v i t y to whether  4.1.5  as  t r a i n i n g times  performance  are  do n o t e a s i l y l e n d t h e m s e l v e s  One c a n a f f i r m ,  however, t h a t  required  indicated. to  a l l subjects  exceeded  during a comfortable t r a i n i n g period extending for is  also j u s t i f i a b l e to  the user  extrapolate  g r o w s more a w a r e o f t h e  i m p o s e d upon h i m .  few  of  performance  to  classification  less  to  66  60 9750% 50 c  (Ul  9500 % 96-25 %  40  •  El  c a  9125 %  93-75 %  9000 %  30  a> >  'a  E o  20  9250 %  9250 %  87-50 %  87-50 %  82-50 %  72-50 %  9000 %  67-50%  4750 %  67-50 %  35-00%  3000  Prof. / Tech.  1  Prof. / Non-Tech.  F i g u r e  4.3  T e s t i n g  Data  R e c o g n i t i o n  f o r  S u b j e c t s  Performance  %  ' Non-Prof./Non-Tech.  Subject category  Whose  A t t a i n e d  95%  10  '  67 4.1.6  Confusion Matrices  L a r g e and s m a l l i n p u t classified,  misclassified  confusion matrices  c h a r a c t e r s were c a t e g o r i z e d as  or u n c l a s s i f i e d .  i n Tables VIII  R e s u l t s appear  a n d I X . Raw d a t a  r i g h t m o s t columns and i m p o r t a n t p e r c e n t a g e s  are  correctly  i n the  form o f  i s summarized i n  shown i n t h e  the  lower righthand  corners. Four subjects to p a r t i c i p a t e  i n the  variance suggested  it  the  i s more l i k e l y  than to  This i s  to  attempt  and  the d i c t i o n a r y "  q u i t e u n l i k e many s y s t e m s  i n which  the  t h e number o f m i s c l a s s i f i c a t i o n s [ C H U I 7 6 J .  dictionary i s very tolerant classification  to u n i n t e n t i o n a l v a r i a t i o n s ,  of a grossly d i s t o r t e d  input  g e n e r a t e an " u n k n o w n " r e s p o n s e . Misclassifications  c e r t a i n members o f t h e i n which d i s c r i m i n a t i o n results  A l o w sample-  conducted.  There were no "unknown t o  reported.  author's  each.  selected  a v e r a g e p e r f o r m a n c e s w e r e 9 6 . 9 and 96.1% f o r l a r g e  number o f unknowns o f t e n e x c e e d s Since  c o n s i s t e n t were  130 c h a r a c t e r s  s u f f i c i e n t t e s t i n g was  small digits respectively. classifications  seemed f a i r l y  t e s t i n g and e n t e r e d  that  The f i n a l  whose r e s u l t s  symbol r e p e r t o i r e . proves poor are  d e p e n d more on u s e r It  usually result  is possible,  rather-ill-defined,i t  to  A major  source o f e r r o r  Classification pen-tip at  is  fairly  a few o f t h e m o r e g e n e r a l  recognition system:  i s i n c o r r e c t pen  i s based on the  posture.  assumption that  i s kept perpendicular to  a l l times.  pairs or  sets clear  that  than on r e c o g n i t i o n s y s t e m l i m i t a t i o n s .  list  encountered by u s e r s u n f a m i l i a r w i t h the (i)  Although character  stroke habits  though.,  from e x c e s s i v e a m b i g u i t y between  the  t h e w r i t i n g medium  Inexperienced users often  have  difficulties  TABLES V I I I  AND I X  CONFUSION MATRIX FOR RECOGNITION PERFORMANCE AFTER T R A I N I N G  TABLE V I I I :  LARGE D I G I T S  (>  5")  SYSTEM C L A S S I F I C A T I O N  0  1 2  3  4  5  6  7  8  9  C  0 16 16  1  16  2 H  16  3 1  4 H  15 13  5 6  U  16  0  0  16  0  0  16  0  0  16  0  0  15  1  0  1  2  13  3  0  16  1  16  0  0  16  0 .0  15  1  16  7  15  1  8  M  0  0 % 36-9 3-1 0  16  9  TABLE I X :  SMALL D I G I T S  16  0  (<|") o  SYSTEM C L A S S I F I C A T I O N  0  1 2  3  4  5  6  7  8  9  C  36 35  1 34  1  1  36 33  2  1  36  0  0  35  1  0  34  2  0  36  0  0  33  3  0  1  34  2  0  35  1  35  1  0  35  1  0  35  1  0  3 5 35  1  0  1  35 1  1  35  % C: M: U:  U  1  34  LEGEND  M  CORRECTLY RECOGNIZED MISCLASSIFIED UNCLASSIFIED  3-9 I96-1  0  69 difficulty  i n complying because of changing  pressures. to  This d i f f i c u l t y  disappear  initial barrel  and c a u s e s  i s u s u a l l y the  first  few p r o b l e m s a f t e r  familiarization period. and v a r i o u s  drag  the  Although the  pen  t r a i n i n g a i d s were d e s i g n e d  f a c i l i t a t e maintaining proper number o f s u g g e s t i o n s  for  pen-tip  angle,  improvement  are  to  a  mentioned  i n Chapter V .  (ii)  Another factor  c o n t r i b u t i n g to observed  classifications  stems from the w r i t e r ' s  to pause or h e s i t a t e explaining position, in  that the  minutes  the  user  a smoother,  mis-  at  stroke  tendency  end-points.  pen s e n s e s p r e s s u r e and i s often motivated to  crisper  fashion.  After  of p r a c t i c e performance  By not  write  several  improves  markedly.  (iii)  A l l testing dictionary  (apart  from the  personalized  evaluation i n Section 4.2)  conducted u s i n g the  restricted  f o r ANSI n u m e r a l s .  Although the  dictionary  i m p l e m e n t e d c l a s s i f i c a t i o n scheme  is  forgiving  regarding  forms are  occasionally misclassified.  p o s s i b l e s o l u t i o n to "generalize" the  number o f  the  spurious  this  inherently  data, non-nominal One  p r o b l e m w o u l d be  dictionary, i . e . ,  character  was  forms  to  to  increase  explicitly  character  70 provided for.  T e s t i n g was c o n d u c t e d w i t h  a d i c t i o n a r y b u t n o a t t e m p t was made t o the  results.  dedicated examined.  Instead  an i n d i v i d u a l  ( i . e . , personalized) S e c t i o n 4.2  investigating  the  is  quantify  user-  s y s t e m was  devoted  salient  such  to  features  o f such  a  proposal.  4-.1.7  Skill  Retention  S i n c e some d e g r e e performance forget  is  attained,  r e c e n t l y mastered  of l e a r n i n g i s  the  q u e s t i o n of whether  These s k i l l s  acceptable  performance  swimming a r e  are  retained  acceptable  the  user  Psychologists talk of  long after  l e v e l has been reached;  system  might  t r a i n i n g has  been  a c q u i r e d by c o n t i n u i n g t r a i n i n g a f t e r  usually overlearned.  pen-training period is  or not  s k i l l s was i n v e s t i g a t e d .  o v e r - l e a r n e d s k i l l s which are completed.  required before  bicycle  The p o s s i b i l i t y  riding  that  the  sufficient for over-learning, w i l l  the  and  40 m i n u t e  be  considered  next. Two seemed at  subjects  who a t t a i n e d  95% a c c e p t a b l e  level  to be p e r f o r m i n g c o n s i s t e n t l y were asked t o undergo  one week i n t e r v a l s  preceded by a b r i e f appears as F i g .  after  the  conclusions the  initial  t r a i n i n g program.  refamiliarization period.  t o be d r a w n .  It  sufficiently may b e n o t e d ,  over-learning condition, i . e . ,  significantly  a n d who  further Each  testing  trial  A summary o f d a t a  was  obtained  4.4.  S a m p l e s i z e was n o t  typify  the  i n the  one month f o l l o w i n g  large  to allow  however,  that results  performance the  general  does n o t  training period.  obtained deteriorate  Immediate  Testing a f t e r  1,2,3  or 4 weeks  A l l T r a i n i n g Sessions Were Preceeded by a 5 Minute R e f a m i l i a r i z a t i o n P e r i o d .  0 Time following F i g u r e 4.4  training  (weeks)  The E f f e c t o f Time F o l l o w i n g T r a i n i n g on R e c o g n i t i o n  Performance  72  A.1.8  Classification  Robustness  A restricted specified  dictionary constructed to c l a s s i f y  i n the p u b l i s h e d ANSI s t a n d a r d  the m a j o r i t y o f e x p e r i m e n t s . ANSI s e t  character  classification  process,  changing the s t a n d a r d  a p p e a r s on t h e output.  nominal the  rather  seemed t o be t o l e r a t e d b y  i . e . , n o n - n o m i n a l forms were r e c o g n i z e d w i t h o u t  for this  inherent  a n d on t h e r i g h t i s  t o l e r a n c e were g i v e n i n  a photograph of  direction-code vector,  i d e n t i t y o f the  character  is.  two w i t h  the  chapter  a pressure  input d i g i t  trace,  a l s o appear  digit  processor's and  i n the  For example, the processor output  slope-code  a n d a p p e a r s as P l a t e  d i r e t t i o n - c o d e are  interfere  with  the  tolerated  Although a pressure  handwritten o r i g i n a l . trace bears pressure  little  i s not  (a)  classification  local variations, it  severe  the  d i r e c t i o n - c o d e v e c t o r p r o v i d e s a d i r e c t measure  distorted  in  the  Occasionally,  i s not Plate  "ULURDLURDRURUO" was of F i g . 4.5.  subject  to  o f how n o n for a nominal  is  (b),  directly related  is  original;  to pen  the  exact  they  velocity.  is  do n o t  unintentional  form o f  the  a case i n p o i n t where it  of  characters.  useful in identifying  F i g . 4.5,  resemblance t o the  correctly  the c o n d i t i o n t h a t  intended to r e f l e c t  A  Non-nominal elements  d e c i s i o n made f o r o t h e r  trace  a  photograph.  two w i t h d i r e c t i o n - c o d e "RDLDRO" i s i l l u s t r a t e d i n F i g . 3 . 7 .  recognized the  left  to the  The  w e r e shown  e v i d e n c e i s p r o v i d e d i n F i g . 4.5 where a handrawn  The g e n e r a t e d  d e c i s i o n as  digit  B e f o r e each s e s s i o n s u b j e c t s  distortions  forms  dictionary.  Explanations Pictorial  character  f o r h a n d p r i n t e d d i g i t s was u s e d  and a s k e d t o c o n f o r m t o t h e s e n o m i n a l s h a p e s .  unintentional  III.  f o r N o n - N o m i n a l Forms  clear that  the  pen-tip  73  Plate  (a)  SLOPE CODE VECTOR ULUROLMRORURUO  PRESSURE INTERPRETATION OF CHARACTER ? *  Plate  (b)  CODE VECTOR DLDRULO;  PRESSURE INTERPRETATION OF CHARACTER S  Processor  Output  Handwritten  Original  Figure  R e c o g n i t i o n Performance w i t h Non-Nominal Input  4,5  74 Complex r e l a t i o n s  have,  l i t e r a t u r e . W. S t a l l i n g s n o t e s by c o n s t a n t changes  angular  writing velocity differs i n s c r i b e d by the the  literature,  hence  the  tolerance  4.2 4.2.1  characters  "Chinese characters  and a v e c t o r "[STAL75] .  u s e r s and a l s o  [MILL69].  In this  velocity  are  trace  author nevertheless  an example o f  are  correctly  in  characteristics.  frequent;  and d i s t o r t i o n .  distortions  the  l i t t l e agreement  Appendix B i s  i n A p p e n d i x B were  that  symbols  performed by the  sample d i s t o r t i o n s  slant  characterized  for different  trial  technical  discovered that  was made t o o p t i m i z e t h e  s u r p r i s i n g l y severe  appearing  c a n be  T h e r e seems t o be  to symbol s i z e ,  i n the  displacement Miller  o f 9 8 . 7 % was a c h i e v e d .  h a s h a d some p r a c t i c e , All  been p o s t u l a t e d  of a 300-character  a p p e a r s as A p p e n d i x A .  system's  between  no e f f o r t  The h a r d - c o p y  the  time  same u s e r  a recognition score  that  acceleration  parabolically with  in fact,  Once t h e  easily  user  accommodated.  classified.  C o n s t r u c t i o n of a P e r s o n a l i z e d System Introduction  Since  a number  characteristics,  of subjects  exhibited irregular  the p o s s i b i l i t y o f p r o v i d i n g u s e r - d e d i c a t e d  s y s t e m s was e x p l o r e d .  In order  personalization process,  Two s u b j e c t s  to develop  performance  system p a r a m e t e r s i n c l u d i n g the were  a c c o r d i n g t o ANSI s t a n d a r d s . t o 90% r a n g e . restrictions  In a second  for  the purpose  Scores were  registered  d i g i t s were  w e r e p l a c e d on c h a r a c t e r  were  investigated. testing.  reasonably w e l l  asked as  to write  falling  a g a i n drawn b u t  formation.  the  various  of r e p e t i t i v e  both performed  The p a r t i c i p a n t s  test,  in  c l a s s i f i c a t i o n d i c t i o n a r y was  chosen  of p r a c t i c e .  recognition  and e v e n t u a l l y a u t o m a t e  s e n s i t i v i t y t o changes  N e i t h e r had been t r a i n e d p r e v i o u s l y yet s e v e r a l minutes  writing  System  digits  i n the  this  after  time  performance  80 no  75 suffered  considerably, possibly for  i)  the  new  'free-form'  the ANSI s e t inherent  ii)  not  a l l system parameters are  Six  experiments  the  f o r the  the  effects  for  are  The v a l i d i t y  the  fall  within  t o be saved by  dictionary,  i.e.,  the  fashion.  following  sections.  In the  In  the  remaining  system parameters are  w e r e assumed  to i n c l u d e the  is  t o be i n d e p e n d e n t  discussed i n chapter  Classification  the influence  investigated. optimal  t e s t i n g had been performed u s e r was r e q u e s t e d C o n s t r a i n i n g the  objective.  V.  o f one  another.  Results  of large  excursions.  Dictionary  to  using a r e s t r i c t e d ,  form a l l c h a r a c t e r s  user  o f the w r i t i n g environment  system design  i.e.,  D e v i a t i o n s o f ±20% h a d a n e g l i g i b l e  f o l l o w i n g pages r e f l e c t  past,  naturalness  were  robust.  t e s t e d was d e f i n e d  assumption  the  pre-specified  being tested,  d i c t i o n a r y tree i s examined.  effects  R e c o n f i g u r i n g the  In  the  colleagues,  d e s c r i b e d i n the  author.  the  performance.  of this  on t h e  subjects  of a l t e r i n g various  on s u b j e c t Parameter  major  reasons:  c l a s s i f i c a t i o n system;  suitable  value determined  the  d i d not  a n d a number o f c o o p e r a t i v e  The r a n g e f o r e a c h p a r a m e t e r  4.2.2  of the  not  sections,  tabulated  characters  and w e r e t o o e x t r e m e  tolerance  restructuring  influence  following  system parameters t h a t had been o p t i m i z e d f o r author  first,  the  t o n o m i n a l forms  and as s u c h i s  fixed  in a nominal, detracts  contrary  to  a  from  76 Crane and Savoie dealt with the problem by " g e n e r a l i z i n g " t h e i r dictionary to accomodate a larger number of character v a r i a t i o n s [CRAN77]. In this thesis the dictionary i s t a i l o r e d to each i n d i v i d u a l user.  Since  a larger generalized dictionary i s inherently more prone to m i s c l a s s i f i c a t i o n s (increased formation  f l e x i b i l i t y for some characters can only be achieved  at  the expense of decreased tolerance for others), the small personalized dictionary could be expected to y i e l d superior r e s u l t s . Detailed stroke patterns of the subjects c a r e f u l l y noted.  In F i g . 4.6  r e s t r i c t e d ANSI form.  r  free-form d i g i t s were  are shown examples of deviations from the  By redesigning the dictionary network and modifying  the associated l i n k e d - l i s t , these otherwise extraneous forms were accommodated.  Results summarized i n the last row of Table X show that  performance for subjects I and I I increased by 42 and 24% r e s p e c t i v e l y . The e n t i r e reconfiguration required approximately one hour of the author s r  time. TABLE X  THE EFFECT OF DICTIONARY RECONFIGURATION  CHARACTER FORM/DICTIONARY  RECOGNITION PERFORMANCE(%) SUBJECT  I  SUBJECT II  Restricted Form / ANSI  85  83  Free Form / ANSI  67  75  Free Form / Reconfigured  95  93  In a non-laboratory optimally reconfigured i n a f a i r l y transparent  implementation, the process of determining  d i c t i o n a r i e s would have to be automated and performed manner.  This problem i s addressed i n Section 4.2.8.  ANSI  SUBJECT  #1  TWO  FIVE  SEVEN  EIGHT  ANSI  SUBJECT  #2  THREE  FIVE  Figure 4 . 6  Examples o f V a r i a n t s o f Nominal ANSI Forms  78  The e f f e c t s performance  are  of other  parameters that  discussed next.  directly influcence  T a b l e X I summarizes  the  data  recognition  obtained.  TABLE X I  SUMMARY OF PARAMETER S E N S I T I V I T Y  TEST  I N A F R E E - F O R M ENVIRONMENT  SEJ^CTED PARAMETER VALUE  DESCRIPTION  PERFORMANCE SUBJECT  -  ANSI D i c t i o n a r y Reconfigured  "  100 R.D.  + Change Z - T h r e s h o l d  R . D . + Change X , Y  "  R . D . + Change Sample  Rate  R . D . + Change C l i p p i n g Time R . D . + Change T i m e - O u t V a l u e R . D . + Change P l a n e Configuration  SUBJECT  71  -  (R.D.)  I  DIFFERENCE(%)*  2 Units 32 U n i t s 0 Units 32 U n i t s 5 Msec. 25 M s e c . 0 Sec 0.2 Sec. 0.2 Sec. 2.0 Sec. Increased H o r i zontal Tolerance. Increased V e r t i cal Tolerance.  II  80  *  100*  87 84 92 66 97 58 71 24  91 91 94 t 40 91 + 75 67 t 43 88 102 67 64  95  95 68 60  A L L RESULTS ARE G I V E N AS PERCENTAGES OF PERFORMANCE WITH A RECONFIGERED DICTIONARY U S I N G GENERAL OPTIMIZED PARAMETERS  f  BOXED VALUES I D E N T I F Y HIGH USER  4.2.3  Threshold  Factors  The i m p o r t a n c e 3.2. Zthe  For input  (i.e.,  SENSITIVITY  data  vertical)  o f t h r e s h o l d p r e s s u r e v a l u e s was e x p l a i n e d i n S e c t i o n  to progress  to subsequent s t a g e s ,  pressure exceed  X - o r Y - component  c e r t a i n minimum v a l u e .  of the  i t was  a pre-specified value.  d r a g p r e s s u r e was a l s o  Predetermined  thresholds  that  The m a g n i t u d e  required  seemed  required  to exceed  adequate f o r  a  most  of  79 subjects.  I n c r e a s i n g o r d e c r e a s i n g thern d r a s t i c a l l y r e s u l t e d  reduction i n system r e c o g n i t i o n  High  thresholds  pressures.  Low t h r e s h o l d s  result  from s t r o k e  end-points)  later  W r i t e r s who h a v e n o t m a s t e r e d spurious  data at Fig.  4.7  Reference-point for  the  author  appears that  stroke is  affect  w r i t e r s who g e n e r a t e l o w w r i t i n g  i n an i n c r e a s e d stages o f the  flow  of data  (especially  c l a s s i f i c a t i o n scheme.  p e n o p e r a t i o n and who, as a r e s u l t ,  end p o i n t s  a performance  parameter  appreciable  performance.  greatly  to  i n an  values  are  s e v e r e l y p e n a l i z e d i n the  graph f o r the  (i.e.,  those  and used i n a l l p r i o r t e s t i n g )  optimal threshold values  are  two s u b j e c t s  chosen as  latter  'generally'  respect  case.  tested.  p r o v i d e d the b e s t  robust with  generate  optimal  scores.  to  It  subject  performance. Performance variations.  high values  to X , Y - threshold  lowered performance one i s  low thresholds  end-point  as  considerably  a heavy w r i t e r .  Neither  they b o t h had l e a r n e d  to  formations.  Sample-Rate Considerations  The m o d i f i e d Early were  s e n s i t i v e , however,  Apparently neither  was h a n d i c a p p e d a t  minimize spurious  4.2.4  quite  In p a r t i c u l a r ,  for both subjects. subject  is  results repeated  decreased reduce  at  the  hardware  data  every  time,  this the  rate.  Since increased data  lowest sampling rate that  r e c o g n i t i o n p e r f o r m a n c e was c h o s e n a n d u t i l i z e d  This parameter  was d e t e r m i n e d  obliging  participants.  test  5 milliseconds.  a c q u i s i t i o n phase i n d i c a t e d t h a t q u a n t i z e d  f o r many c y c l e s a t  system response  sampled pen pressure  from r e s u l t s  for the  author  and a  meant  did  i n general  data  not testing.  few  80  Reference  CO  c  o o  4=  0)  <U V)  a  Level  et_  O CO  x:  O CO  ©  0  o c  G) O  c  a  O  £ o l_ (1)  CL  o UD  O LO  O  z  0  5  32  2  5  32  Pressure  Legend :  F i g u r e 4.7  The  threshold  ( U n i t s of  Triangle  -  Subject  I  Circle  -  Subject  II  Shaded  -  x-,y-Thresholds  Blank  -  z-Threshold  E f f e c t of Changing P r e - s e t  Thresholds  2"^ v o l t )  Pressure  81  In general, writer  should not  end-points the other  are  a h i g h sample r a t e s u i t s  encounter  any d i f f i c u l t i e s  r e l a t i v e l y free  hand,  could expect  a fast  at  from s p u r i o u s  writer.  But a slower  this higher rate i f  data.  to p e r f o r m p o o r l y at  The f a s t e r  writer,  a l o w sample r a t e  a large proportion of h i s  s h o r t - l i v e d quantized d i r e c t i o n s would  truncated  algorithm.  by the  clipping  Experimental results in  Fig. 4.8.  inadequately Again,  the  Subject at  a l o w r a t e and i s  4.2.5  Time  Clipping  Hand-printed  characters  design goals  a high rate  undoubtedly  the  clipping  relatively long-lived  plotted  performs writer of  the  two.  points  often exhibit unintentional end-point  a l g o r i t h m was a p p l i e d a f t e r data.  formations  resolve to  the  No i n f o r m a t i o n i s  since data  would p r o b a b l y e x p e r i e n c e end-point  stages.  c l i p p i n g values  some d i f f i c u l t y  since  at  at  the  case  following  writer  b o t h ends o f each  A fairly  to p r o g r e s s  c o u l d be e x p e c t e d  p r o p o r t i o n of the  careless  low c l i p p i n g values  d a t a w o u l d be more l i k e l y  The f a s t e r  i n the  testing  same q u a n t i z e d d i r e c t i o n .  w a s f o u n d t o b e o p t i m u m f o r most s y s t e m u s e r s .  unintentional  data.  information  threshold  lost  immediately  A v a l u e o f 15 m i l l i s e c o n d s o f c l i p p i n g  analysis  faster  and a r e  extraction.  to discard s h o r t - l i v e d end-point  truncated  at  since  be  o f T a b l e I w a s t o e l i m i n a t e any s p u r i o u s  beginning feature An e n d - p o i n t  of  above e x p e c t a t i o n s  on  b e s t p e r f o r m a n c e was s c o r e d u s i n g t h e p r e v i o u s l y c h o s e n values.  before  the  I who p e r f o r m s w e l l  reference-point  One o f t h e  support  stroke  total  to  character  writer since  to  suffer that  stroke  subsequent at is  high clipped  82  cn c CD  a £ i_  o c  Reference  T-l O O  <X> Level  o cn o co  a> o c o  o  a.  o  £  A O  o  IT)  O  o  A Sample p e r i o d ; 25  Clipping  time:  o  Legend  F i g u r e 4.8  15 0.015  5 0.25  (sees.)  Triangle  Subject I  Circle  Subject I I  Shaded  Sample Rate  Blank  C l i p p i n g Time  The E f f e c t o f V a r y i n g Sample Rate and C l i p p i n g Time  (msec.)  might  rise  to the  information  is  point  also  that  a s i g n i f i c a n t amount  of nominal  discarded.  Data p l o t t e d i n F i g . 4 . 8 concur with expectations Sample r a t e fairly  a n a l y s i s of  quickly;  reference  level)  are  equally affected  equally  4.2.6  as  the  i n the  this point  skilled  at  h i s performance amount o f d a t a  falls  appreciably  clipped rises.  low c l i p p i n g s i t u a t i o n .  This i s  subjects  reasonable  t h e y have h a d i d e n t i c a l t r a i n i n g and a r e  avoiding unintentional end-point  writes  2 4 %of  (to  The  I  above.  therefore  data.  Time-Out Value  The characters;  t i m e - o u t v a l u e was i n t r o d u c e d t o a c c o m o d a t e  effects  If  the  of non-normal s e t t i n g s  time-out  value i s  strokes  The  "S", f o r example i s  ANSI d i g i t ,  values  the  Tipper  a p o o r l y drawn  rightward stroke  it's  decision until Results  d r a m a t i c a l l y the  i g n o r e d and t h e  user  are not  finds  detected.  at  low time-out  recognizer  postulates  the  time i s very slow s i n c e  subsets of others  sytem r e c o g n i z e s the  from e x p e r i m e n t a t i o n c o n f i r m less  perturbed  for  the p r o c e s s o r must  reserve  time-out condition.  these g e n e r a l i z a t i o n s .  by a low t i m e - o u t  value;  p e n a l i z e d o n m u l t i p l e - s t r o k e s y m b o l s and p e r f o r m a n c e  accordingly.  here.  a two-stroke symbol but  system response  whose s t r o k e s ' are  writer is  is  mu3.tiple-stroke  eight.  characters  faster  analyzed  of multiple-stroke characters  At h i g h values  final  are  shortened  that successive  is  discussed  the p r e v i o u s s e c t i o n i n d i c a t e d t h a t S u b j e c t  consequently  its  s i n c e up u n t i l  character  the  slower  scores  fall  The  writer  84  Both writers commented that response time was case of high values.  i n t o l e r a b l e i n the  As expected the slower writer's performance improved  even beyond the reference l e v e l since previously m i s - c l a s s i f i e d multiplestroke symbols were now Fig. 4.2.7  classified correctly.  Results are i l l u s t r a t e d i n  4.9. Plane Non-Uniformity  Quantization  plane rotation and quantization non-uniformity were  mentioned i n Section 3.3.2  when discussing feature extraction c r i t e r i a .  Results of t e s t i n g the e f f e c t s of plane rotation suggested that the performance of most subjects was abruptly.  f a i r l y constant  up u n t i l ±15°  rotation then dropped  The e f f e c t s of quantization non-uniformity were l e s s consistent  and are therefore analyzed here i n d e t a i l . By enlarging the horizontal quadrants, characters with h o r i z o n t a l l i n e s drawn poorly are more often accomodated. three drawn as "d"  , would be recognized  For example, the ANSI d i g i t  correctly.  Similarly, variations  i n the v e r t i c a l d i r e c t i o n are tolerated by enlarging the upper and lower quantization segments. acceptable  to t h i s new  The ANSI d i g i t four drawn as, "s^ , 1  would be  c l a s s i f i c a t i o n system.  Results are i l l u s t r a t e d i n F i g . 4.9.  Subjects  I and II are  affected equally by the non-linear planes and neither the horizontal nor v e r t i c a l quadrants i s favored. d i g i t samples i n F i g . 4.10  enlarged  Observing the handwritten  reveals the reason for t h i s :  characters of  neither subject are slanted excessively. Again the reference l e v e l parameter optimized ( i n this case a zero non-uniformity factor) was for both p a r t i c i p a n t s .  for the author  found to function w e l l  85  O)  c  »  Reference  o o T-l  Level —*  o E t_ o c  a) o c o E  o cn  o co  O  A O  0 A  o LO  Q.  O  O -4-  Time-out Tolerance  short (-2) horizontal  Legend  long ( 2 sec.) vertical  n o r m a l (-6) normal  Triangle -  Subject I  Circle  Subject I I  Shaded  Time-Out  Blank  Stroke  Value  Direction  Tolerance  F i g u r e 4.9  The E f f e c t of V a r y i n g Time-Out Value and Plane C o n f i g u r a t i o n  86  Specimen I  I £  3  / Q  S .4-  (k  H- ^  ^ ^ 7-  & X  9  Specimen I I  I Z 3 4- 6- £ / /2 / 3  F i g u r e 4.10  ;V  -7  3 ^ jT^f 7 3 ^ iT^T 7 • 5*  .9  ^  Specimens of S u b j e c t s ' I and I I Handwriting  87 4.2.8  Automating the  In performance  the  P e r s o n a l i z a t i o n Process  course  o f the  achieved u s i n g parameter  o n l y exceeded  on one  occasion.  optimal for both subjects. are  user  for a l l users.  dedicated  Parameters  It  For this  to t h e i r  tolerant  reason  totally  the  optimal defined  author author  was proved  system  upon the dedicated  set  parameters  to accept  it  parameter  constraints  i n the  was d e v i s e d .  normal fashion,  values.  This information i s retrieved  the  then  performance,  processor  decides user-  under a  user  periods.  was i m p l e m e n t e d u n d e r  microcomputer environment  the  optimal  stored  i n subsequent work  p e r s o n a l i z a t i o n process  discussed e a r l i e r  the i n the  theses.  Parameter  o p t i m i z a t i o n was a c h i e v e d t h r o u g h m o d i f i c a t i o n o f  values  the  as  had been p r e - d e f i n e d , performance  trial  the  proceeded.  processor  Since the  was a b l e  for each of s e v e r a l parameter  b e s t v a l u e s were  the  environment.  After  d i c t i o n a r y c o n f i g u r a t i o n and s e l e c t s  number a n d i s  o f the  i n a non-laboratory  c o n s i d e r a t i o n was n o n - t e c h n i c a l u s e r  been i n p u t  A u t o m a t i o n o f the  dictionaries  personalization i s worthwhile.  most s u i t a b l e  identification  who h a d  acceptable  p e r s o n a l w r i t i n g s t y l e s w e r e more c o m f o r t a b l e w i t h  alone,  has  dictionary provided  two s u b j e c t s  transparent p e r s o n a l i z a t i o n process  character  input  to determine  values.  character  set  subject  In t h i s way, a set  of  selected.  A method evident.  optimized for  restricted  However, the  S i n c e a major  parameter  optimized for the  appears that  s y s t e m a n d w o u l d b e more l i k e l y  a  values  reference-level  independent. The i n h e r e n t l y  results  12 p e r s o n a l i z a t i o n t r i a l s ,  of automatic  A ten-level six-link  p e r s o n a l i z e d d i c t i o n a r y g e n e r a t i o n was p e r node  tree  requires  i n excess  of  one  less  megabyte  o f memory; c o r e  requirements  a p p l y i n g such a technique. used i n p r e v i o u s the  final  Furthermore,  the hand-configured  t e s t i n g was o p t i m i z e d a t  form r e q u i r e d under 250 w o r d s .  efficiently  perform t h i s  An a l t e r n a t i v e subjects  a l o n e e l i m i n a t e d any p o s s i b i l i t y  intermediate  it  became  success.  clear that  c o u l d be c a t e g o r i z e d i n t o a s m a l l number o f g r o u p s . in  stored  dictionaries  were i n t e r r o g a t e d  values  were  reports  subject  modified.  general  purposes.  subject  at  is  user  i l l u s t r a t e d i n F i g . 4.11.  at  sentiments  4.3  4.3.1  that  s h a l l be  Performance w i t h  Effects  In  re-iterated  an E x t e n d e d R e s t r i c t e d  section  the  number o f s p e c i a l c h a r a c t e r s . applicable  since  and  and  demonstration  s e s s i o n end at  (i.e.,  after  the bottom o f  interdependence  reflects  i n Chapter V .  upon R e c o g n i t i o n o f t h e  this  task  parameter  parameters  The s t a t e m e n t  s c r e e n w h i c h warns o f p o s s i b l e parameter  and  session.  d i c t i o n a r i e s was d e v e l o p e d f o r  I n f o r m a t i o n r e l a y e d to the  10 m i n u t e s )  constructed  successively. Simultaneously,  s y s t e m v a r y i n g two n o n - r o b u s t  two d i f f e r e n t  or  performed a p r e d e f i n e d w r i t i n g  the end o f the w r i t i n g  styles  trends  The p r o c e s s o r m o n i t o r s r e c o g n i t i o n a c c u r a c y  performance  A prototype referencing  As the  observing  their writing  M o d e l d i c t i o n a r i e s o f 200 t o 300 w o r d s w e r e  i n disk files.  to  After  These  process  effort.  s y m b o l f o r m a t i o n may b e r e g i o n a l ( S U E N 7 9 ) , p r o f e s s i o n - r e l a t e d  psychological.  the  stages o f the  considerable  scheme was e m p l o y e d w i t h  over a p e r i o d o f months,  dictionary  D e s i g n i n g an a l g o r i t h m  function would require  of  Dictionary  Original  Character  ANSI r e p e r t o i r e  i s extended  The a d j e c t i v e  'restricted'  only nominal character  forms are  Set  to include a is  explicitly  still defined i n  the  SUBJECT  NAME  ,. S T E V E W I L L E Y  DICTIONARY* RECOGNITION PERFORMANCE AT 5 MS SAMPLE RATE RECOGNITION PERFORMANCE AT^ SS MS SAMPLE RATE RECOGNITION PERFORMANCE AT LOW ORAO THRESHOLDS RECOGNITION PERFORMANCE AT HIGH DRAG THRESHOLDS  I;  -.7 i,B  <B >,B  DICTIONARY i RECOGNITION PERFORMANCE AT S MS SAMPLE RATE  -.9  RECOGNITION PERFORMANCE AT IS MS SAMPLE RATE  iJ  RECOGNITION PERFORMANCE AT LOW DRAG THRESHOLDS RECOGNITION PERFORMANCE AT HIGH DRAG THRESHOLDS  ..3 >.3  CHOICE OF OJCTJONAUr AMD OPTIMIZED *Y$TEM VARIAILtt I t NOT UWtVlKKT, IxturricltKT FAffAHETEff INTERDEPENDENCE INFORMATION.  *  F i g u r e 4.11  Example o f D i s p l a y e d Output from t h e Automated P e r s o n a l i z a t i o n Scheme  „  90 dictionary.  The  'original  used i n a l l p r e v i o u s  character  testing.  r e c o g n i t i o n performance  are  their  practice  average  score  f o r 5 minutes  examined  refamiliarization  discussed e a r l i e r .  f o r use  was v a g u e l y  seen  the  times,  the  size  trained  to  95% l e v e l w e r e  During this  graphics  upon  the  editor  part  some o f t h e m b u t since  and as a r e s u l t  allowed  set  implementation  experienced  no  the  judged order  their of m u l t i p l e  was o b l i g e d t o  person write  usual.  o f the  test,  subjects  the numerics  special symbol-set.  or w r i t i n g medium.  4 . 1 2 and e x p e r i m e n t a l  point  period of  The n o n - p r o f e s s i o n a l / n o n - t e c h n i c a l  than  set  professional/non-technical  anticipated  random f a s h i o n ,  introduced 9-character  Fig.  than  symbols before  first  in a fairly  character  the  formal t e s t i n g .  familiar with  s l o w e r a n d more h e s i t a n t l y In  exceeded  new s y m b o l s ; t h e  h a d t o be m e m o r i z e d .  had never  who h a d p r e v i o u s l y b e e n  i n the d e d i c a t e d  c o n s t r u c t i o n more d i f f i c u l t strokes  of d i c t i o n a r y enlargement  The p r o f e s s i o n a l / t e c h n i c a l s u b j e c t  i n forming the  participant  ANSI d i g i t  e a c h p a r t i c i p a n t was e x p o s e d t o a 9 - s y m b o l c h a r a c t e r  w h i c h was d e s i g n e d  difficulty  to  the h a n d p r i n t e d  here.  consistently  prior  is  The e f f e c t s  T h e same t h r e e s u b j e c t s where  set'  results  The t e s t are  were  set  asked to w r i t e  plus  the  No c o n s t r a i n t s s c o r i n g form i s  four  recently  were p l a c e d shown  summarized i n Table X I I .  in  on  to  GENERAL  DICTIONARY 1  CONSTRAINTS System  Test  PLACED  TESTING UPON T H E USER  Parameters  SU3JECT  SUBJECT  CATEGORY  IDENTIFIER  Writing  Speed:  none  Medium:  none  X,Y-  Pressure  Threshold:  Tine-out  value  Z  Pressure  Threshold:  C l ! p p i ng  Time  -  CHARACTER  Writing  1  DIGIT C  Not  M  (sec.):  C  b)zero  as 0  (deg's.):  N U M B E R  3  M  conforming  Non-1inearizotlon  2 U  a)ANSI  Plane  S E S S -1 O N  TEST  SIZE  Stylistic: ( s e c ) :  U  C  M  • U  l|  c  M  • U  zero  Specified  one two three four five six seven eight nine  i  Sub-Totals Resistor Capacitor Inductor F.E.T. V.  Source  C.  Source DELETE  ! 1  REVERSE  (  FINISH  Sub-Totals j I  1  i  1  M  I  ^  I 1  TOTALS  Legend: . • * con forming to ANSI standards for handprinted characters  F i g u r e 4.12  T e s t i n g w i t h an Extended, R e s t r i c t e d  C - classified correctly M •» ."nisclassJfled li - unknown to the dictionary  Dictionary  92  TABLE X I I SUMMARY OF A N S I VERSUS EXTENDED DICTIONARY TEST RESULTS  CATEGORY/ A V G . PREVIOUS A N S I SUBJECT D I G I T SCORE (%)  I N I T I A L AVG EXTENDED DICTIONARY SCORE (%) ENTIRE  D I G I T S ONLY Prof./ Tech. P r o f . /NonTech. Non-Prof. Non-Tech  are  was s u s p e c t e d classification  95.00  92.10  2  95.00  92.50  84.20  10  97.50  96.25  85.52  10  very nature  accommodated that  at  the  of the  dictionary's  expense  o f reduced  additional character  performance.  the  extended  concerning the performance  tests  of d i c t i o n a r y enlargement.  f o r each subject  using different  with  it  affect the  and by r e p e a t i n g t o draw  new  freedom i n o t h e r s  forms might a d v e r s e l y  d i c t i o n a r y i t was p o s s i b l e  effects  construction  Data f o r d i g i t performance  d i c t i o n a r y was a v a i l a b l e from p r e v i o u s under  DICTIONARY  96.25  S i n c e by the characters  •ENTIRE DICTIONARY ADDITIONAL TRAINING TO 95% L E V E L ( M I N . )  the  ANSI trials  conclusions  Differences  in  mean  d i c t i o n a r i e s are a n a l y z e d  below. The s a m p l e i s thesis, applied.  the  Student's  t  p a i r e d and f o r  reasons g i v e n i n S e c t i o n 4 . 1 of  d i s t r i b u t i o n with n-1  degrees of  freedom w i l l  this be  93  TABLE  HYPOTHESIS  TESTING:  XIII  VARYING THE DICTIONARY  OBSERVED PERFORMANCE FOR D I G I T S ONLY (%)  CATEGORY/ SUBJECT  (ANSI)  X  ±  D = (X^ - X )  X (EXTEND) 2  2  (D - D)  Prof./Tech.  95.00  95.00  0.00  Prof./Non-Tech.  95.00  92.50  2.50  Non-Prof./ Non-Tech.  97.50  96.25  1.25  D =  the  score  Since the  average  under  extended  the  sample  d i g i t score  dictionary,  p o p u l a t i o n parameter  -  1.250  1.563  0.000  0.000  -  1.25  for  s  the  a one-sided  testing  u n d e r ANSI and E x t e n d e d d i c t i o n a r i e s be r e p r e s e n t e d  A > -  Since A = 0 i s recognition affected  5% s i g n i f i c a n c e  4.3.2  Effects  i n c l u d e d i n the  less  2 degrees o f freedom  confidence  i n t e r v a l , the  is  subject by A .  Then,  (cf.eqn.(5))  null hypothesis,  for previously e x i s t i n g characters  defined dictionary expansion,  is  acceptable  i s not at  upon A v e r a g e D i c t i o n a r y C l a s s i f i c a t i o n  for  the  encouraging;  entire it  is  extended  fairly  character  the  classical  Performance  set  was,  that  adversely  level.  Performance initially,  with  in  exceeds  0.85  performance  by t h e  (—) /-—I vn  difference  procedure  performance  . UJ  for the  =1.563  2 D  ANSI d i c t i o n a r y  L e t the  t  2  1.563  1.250  preferred.  A > D -  (D-D)  (DIFFERENCE)  at  least  obvious i n examining Table X I I  94  that  difficulties  were  encountered  with  the  new s y m b o l s e t .  t e c h n i c a l p e r s o n who h a d h a d p r e v i o u s e x p e r i e n c e suffered light  less  than  his  of sentiments  symbols  are  fellow  expressed  inter-category b i a s e d sample  it  is  apply only  plus  special characters  experience  deform the  to omit t h i s  f o r those subjects the  i s not  s p e c i a l symbols surprising in  the  familiar  and u s e d t h a n u n f a m i l i a r s y m b o l s .  Rather than  of  This result  previous  justifiable  Calculations digits  this  variance.  drawing the  b y p s y c h o l o g i s t s who c l a i m t h a t  more e a s i l y u n d e r s t o o d  Unfortunately,  results  subjects.  The p r o f e s s i o n a l /  the  data.  With  large  by i n c l u d i n g  a  this modification  the  i n performance extended  i n a rather  statistic  unfamiliar with  difference with  results  new s y m b o l s .  for d i g i t s only  d i c t i o n a r y are  and  indicated  in  Table X I V .  TABLE X I V HYPOTHESIS T E S T I N G : . VARYING THE SYMBOL SET (Subjects u n f a m i l i a r w i t h extended symbols)  OBSERVED PERFORMANCE (%) D = ( X  CATEGORY/ SUBJECT  X  x  (DIGITS)  X  (ENTIRE)  2  x  - X ) ?  (D - D)  (D -  D)  (DIFFERENCE)  Prof./Non-Tech.  92.50  84.20  8.30  -1.215  1.48  Non-Prof./ Non-Tech.  96.25  85.52  10.73  1.215  1.48  B The  p o p u l a t i o n parameter,  A > D -  A > 1.83  t  A, which  n  at  represents  (~T;) w i t h n the  2  =  this  1 degree o f  5% s i g n i f i c a n c e  9.515  difference  freedom  level  S  2 d  =2.96  i s bounded  as,  A = 0 i s not  Since interval,  the  performance  "no d i f f e r e n c e " h y p o t h e s i s  of subjects  performance w i t h This tolerant  to  i n c l u d e d i n the  unfamiliar with  result  users'  Each s u b j e c t  seemed  to imply  95% l e v e l  to t e s t  speed  95% a f t e r  (hence  It  is  their  and n o t  the  dictionary. insufficiently A  final  special character  forms  the p r e v i o u s l y a t t a i n e d  3 minute  difficulties  problems  training period.  stemmed f r o m a  acquired)  each a t t a i n e d  The  other  reduced  d i c t a t e d by the  95%  their  acceptable  difficulties,  a l l  level  of  subjects  set.  concluded, therefore,  t h e new t a s k  to  possibility.  that  the  r e c o g n i t i o n p e r f o r m a n c e was d i r e c t l y r e l a t e d with  same  i.e.,  inferior  d i c t i o n a r y was  Despite i n i t i a l  t o t h e new s y m b o l  5% l e v e l ,  quickly eliminated his  more g o o d a n d b a d d a t a  of practice.  the  from n o m i n a l forms.  this  a brief  t h e new s y m b o l s e t ;  10 m i n u t e s  adapted  the  had improved to  after  discovered that  unfamiliarity with  quickly  that  The p r o f e s s i o n a l / t e c h n i c a l s u b j e c t  two s u b j e c t s  at  confidence  s p e c i a l symbols i s  was a s k e d t o p r a c t i c e w r i t i n g  and a c h i e v e d t h e  writing  the  unintentional deviations  r e c o g n i t i o n performance  level.  rejected  common n u m e r i c s e v e n w h e n u s i n g t h e  e x p e r i m e n t was d e f i n e d i n o r d e r  until  is  5% o n e - s i d e d  initial  to  the  t o any i n h e r e n t w e a k n e s s  reduction in  subjects' of  unfamiliarity  the c l a s s i f i c a t i o n  scheme. It particular  must be k e p t  to  the  described experiment,  f r o m 1 0 t o 19 c h a r a c t e r s , character  sets.  performance  i n mind, however,  i.e.,  the  these r e s u l t s  by r e d u c i n g the  p e r s o n a l i z a t i o n s h o u l d improve  This i s  a l l o w e d some f r e e d o m a n d f l e x i b i l i t y  e v e n more t r u e i f t h e  i n symbol  definition.  larger  system  t o t a l n u m b e r o f f o r m s t h a t must be  dictionary.  are  enlarging of a dictionary  and a r e n o t n e c e s s a r i l y a p p l i c a b l e t o  Nonetheless,  accommodated by the  that  explicitly designer  is  96  V.  5.1  Summary o f E x p e r i m e n t a l R e s u l t s  Testing with of  CONCLUDING REMARKS  15 s u b j e c t s  subjects  the  ANSI r e s t r i c t e d  and r e s u l t s  were  later  characters  were  previously  trained  for  selected  obtained.  d i c t i o n a r y was c o n d u c t e d  o v e r 4000 i n p u t  characters  during personalization t r i a l s  F i n a l l y , t h e extended  subjects  were  symbol set  on a  total  gathered.  Two  and a f u r t h e r  1000  was t e s t e d  on 3  and a n a d d i t i o n a l 4 0 0 r e c o g n i t i o n d e c i s i o n s  were  noted. The d e s i g n p r o v i d e s communication ,  a major  objective  d i s t r a c t e d by o p e r a t i o n a l  task w i t h  automated  subjects  t h e i r normal w r i t i n g Owing t o  positively all  were  the  s m a l l amount  that  character  and t h a t  they  d i d not  set  as  occasionally  to  at  to  s e s s i o n was just  the  fully  s l i g h t l y below  i n the  of hardware  pen-paper  were  acceptable,  from the n a t u r a l n e s s  of the  results.  unintentional end-point  generate spurious  classifications.  easily  and  they  understood,  system.  i n a situation  The b r i e f data  and  They w e r e  improving recognition accuracy scores  the p e n w o u l d be most s u i t a b l e  reduce  involved,  environment  training continued.  constraints  detract  seems t h a t  user  The t e s t i n g  and s i m p l i c i t y  u s e r s had p r e v i o u s l y a c h i e v e d a c c e p t a b l e the  not  to a p p l y t h e m s e l v e s  to enter d i g i t s  comfortable  more c o n f i d e n t  felt  allows  man-machine  S i n c e u s e r s were  they were able  able  for  rate.  m o t i v a t e d by t h e i r  It  thesis.  apparent ease.  even n o n - t e c h n i c a l u s e r s were grew p r o g r e s s i v e l y  environment  of this  factors  designated w r i t i n g a n d most  a natural  that  where  training period might  97 During t e s t i n g , and t h e  operator  any i n i t i a l  progressed  d i f f i c u l t i e s were  i n approximate  e x p o n e n t i a l l e a r n i n g curve s t a t i s t i c s , 90% a f t e r  15 m i n u t e s  of  accordance w i t h  attaining  identical. analysis  S e n s i t i v i t y to  Inferences  in  that  almost  proposed  after  calculations  of user  15 m i n u t e s  subject  were c o n c l u s i v e :  was i n d e p e n d e n t  classical of  training.  performance  of variance  overcome  a recognition accuracy  Although t e c h n i c a l users performed b e t t e r users i n i t i a l l y ,  quickly  non-technical  o f p r a c t i c e was  almost  c a t e g o r y was c a r e f u l l y a n a l y z e d b y a p p l y i n g to the  after  category.  a l l u s e r s are  than  large  15 m i n u t e s  quantity  of data  of t r a i n i n g , user  obtained. performance  A primary system design o b j e c t i v e  capable of  communicating n a t u r a l l y  is  under  met the  environment.  3 The h y p o t h e s i s correctly  recognized than  independence Objective  the  need  to competently  obtained  training session. groups  for  performance training  t h a n —")  was e x a m i n e d .  Since  are  more  of user  this  performance I,  f e a t u r e was i m p o r t a n t .  f o r each  user  category  d u r i n g and a f t e r  The s t a t e d h y p o t h e s i s was r e j e c t e d  group,  after  reduction  the p r a c t i c e  d u r i n g the  for  t r a i n i n g p e r i o d but  15 m i n u t e s  i n performance  session.  was s h o r t - l i v e d a n d ,  session.  test  often  A  all  the three  ( P r o f e s s i o n a l / T e c h n i c a l , P r o f e s s i o n a l / N o n - T e c h n i c a . l and  A slight recorded  (taller  performed.  Non-Professional/Non-technical) independent  characters  small characters  a n a l y s i s was a g a i n  Data were  category  large  t o s y m b o l s i z e was a l s o a d e s i g n s p e c i f i c a t i o n ( T a b l e  #3),  statistical  brief  that  of for  Fortunately  in fact,  Examples o f very l a r g e  accepted,  training. small characters this  difference  non-existant  at  the  was in  end o f  and v e r y s m a l l c o r r e c t l y  the  classified  digits  appear It  as A p p e n d i x B . i s perhaps  worth noting that  large  characters  1 cm. o r e v e n 10 c m . i n h e i g h t . B e c a u s e  of the  chosen  scheme,  fairly  absolute.  character Left-  s i z e independence  preferred  was f o u n d t o b e a p r e - r e q u i s i t e  and s i n c e t h e  suggested  limited  handed p e o p l e were compared.  During tested  were not  cited  i n current  the able  left-handed  the  the  c l o s e l y resembles  initial  short  to achieve  literature  for  R e c o g n i t i o n performance  u s e r was r i g h t - o r  [MILL69,  that  experience  left-  was n o t  tasks.  and. r i g h t found to  be  left-handed.  training period,  the  Since  good r e c o g n i t i o n  people might  t r a i n i n g sessions  to  extraction  and c o n s e q u e n t l y p e r f o r m p o o r l y i n w r i t i n g  Data o b t a i n e d d u r i n g and a f t e r  to whether  for  Xebec pen posture  by a r i g h t - h a n d e d p e r s o n ,  discomfort or d i f f i c u l t y  senstive  feature  not  and r i g h t - h a n d e d p e o p l e h o l d t h e i r pens d i f f e r e n t l y .  c o r r e c t pen posture performance  is  are  acceptable CHUI76].  the m a j o r i t y o f  recognition level Two q u e s t i o n s  subjects  o f 95%  thus  raised  were: (i)  Could  the  acceptable  under the design?  (ii)  And i f  attaining  Test p a r t i c i p a n t s allowed Fig.  subjects  were  t o p r a c t i c e beyond the  4.3 express tested  Furthermore,  performance attained  the  achieved  e x i s t i n g hardware/software  average  the  user have  n o r m a l 15 m i n u t e l i m i t .  l e v e l s at  to  95% p e r f o r m a n c e  chosen from d i f f e r e n t  practice  level?  user  categories  Results  less  than  one hour o f  always improved w i t h  and  appearing  various stages of t r a i n i n g .  95% l e v e l w i t h  r e c o g n i t i o n performance  system  so,  How l o n g w o u l d t h e before  l e v e l e v e r be  further  in  A l l practice. practice.  Appendix A i l l u s t r a t e s o v e r a l l performance Skeptics on  the  part  of the  a 300 c h a r a c t e r  exceeds o f the  s y s t e m c l a i m e d t h a t a n y scheme  operator  were  asked  was i n h e r e n t l y  to return  at  Fig.  4.4.  one m o n t h p e r i o d f o l l o w i n g classified  as  a "once  To t e s t  task;  test  initial  learned  training.  r e c o g n i t i o n performance.  f a m i l i a r but  Scores  T h e c o n f i g u r e d d i c t i o n a r y was f a i r l y explicitly  accommodated  Observed t o l e r a n c e structure the  than  to  deviations  local  to content.  decision repertoire,  it  and d i s t o r t i o n s .  they  illustrated  in  Xebec pen c o u l d  over-learned)  the be  skill.  i n order  to  9 6 . 5 % f o r u s e r s who w e r e  the ANSI n u m e r i c s i n that  d e v i a t i o n s was t h e r e f o r e  the  character  number  more due  an " U n k n o w n " r e s p o n s e  to  is  Examples of a c c e p t a b l e  to  set.  of  low. dictionary  i n c l u d e d in.  never occured during confusion m a t r i x  T h e r e c o g n i t i o n scheme w a s f o u n d t o b e v e r y t o l e r a n t variations  session  from n o m i n a l forms was q u i t e  Although  users  hypothesis,  are  constructed  restricted  learning  d i f f e r markedly over  (i.e.,  averaged  only s l i g h t l y experienced with  this  A t each  U s i n g the  always r e t a i n e d "  requiring  results  occasional users d i d not  C o n v e n t i o n a l c o n f u s i o n m a t r i c e s were quantify  i n which  i n that occasional  one week i n t e r v a l s .  asked to perform a s m a l l w r i t i n g for  deficient  operating.  were  Performance  trial  98%.  w o u l d s p e n d more t i m e t r a i n i n g t h a n subjects  recognition accuracy  testing.  unintentional  distortions  are  i n F i g . 4 . 6 and i n A p p e n d i x B .  illustrated  A few s u b j e c t s and c o m p l a i n e d t h a t t h e environment.  found the constraints  For these people,  explicitly  accommodated  retesting,  their  ANSI n u m e r i c s detracted  dedicated  set  awkward t o  conform  f r o m an o t h e r w i s e  natural  or p e r s o n a l i z e d d i c t i o n a r i e s  t h e i r n o n - n o m i n a l ANSI forms were  r e c o g n i t i o n performance  configured.  i m p r o v e d b y an a v e r a g e  of  to  that Upon  almost  100 A total their  effects  of s i x s y s t e m p a r a m e t e r s were i n d i v i d u a l l y  upon r e c o g n i t i o n p e r f o r m a n c e .  severely influenced testing of  u s e r s were  scores,  is  a desirable  or c a r e l e s s ,  q u a l i t y i n any  f o r example)  the n o n - t e c h n i c a l user  cannot  p e r s o n a l i z a t i o n process to  the  user.  in  designing this  can b e n e f i t  o f the  often  group  tested.  system. characteristics  from parameter  be e x p e c t e d  has been f u l l y  Transparency  values  p a r a m e t e r s o p t i m i z e d f o r a. s m a l l g r o u p  Some u s e r s who e x h i b i t e x c e p t i o n a l w r i t i n g fast  for  A l t h o u g h extreme  f o u n d t o be o p t i m a l f o r a l l s u b j e c t s  Such robustness  examined  fine-tuning.  automated  and p r o c e e d s  o f t e c h n i c a l d e t a i l s was a p r i m e  almost  at  the  is  symbol d e f i n i t i o n .  t o the  tend  f o r the  the  time  u s e r s may b e s a t i s f i e d w i t h  any i n f l u e n c e the  designed  were  the  obtained taken  user  f u n c t i o n pads  they  or  ANSI n u m e r i c s  set  Although  can be  of d i c t i o n a r y c o n s t r u c t i o n .  a standard  on u s e r s '  interactive  dedicated  H o w e v e r , many  plus  8 to  and t e s t e d u n d e r  increase for  the  i f doubling symbol s e t  r e c o g n i t i o n performance.  graphics  10  A 19-symbol set  analysis  of  had  intended was  conditions. r e s u l t i n g performance  i n d i c t i o n a r y s i z e had no e f f e c t o r i g i n a l numeric character  o v e r 19 c h a r a c t e r s )  size  implementation described i n Chapter I I  simulated  set.  d i c t i o n a r y performance  to  reconfiguring dictionary  re-definable  conducted to determine  A statistical that  constructed  related  characters. Tests  for  unknown  consideration  q u a l i f i e d computer programmer.  time symbols are not  u s e r ' s needs at  special  Specially  to be e x p e n s i v e whereas  a simple task  present  the  system.  extended keyboards software  Since  to perform system m o d i f i c a t i o n s  One o f many a d v a n t a g e s t h e p e n h a s o v e r t h e k e y b o a r d i s non-standard  (light,  data  revealed  on r e c o g n i t i o n  scores  N o r was a v e r a g e  (i.e.,  influenced,  although  101  scores  did f a l l  of  s p e c i a l symbol  the  5.2  initially  accustomed  to  drawing  members  set.  Achieved Design Objectives  With the available, design in  s i n c e u s e r s were not  i t was p o s s i b l e  objectives.  Table I  with  r e c o g n i t i o n scheme  are  to e v a l u a t e  In the paragraphs  discussed  completed the  and p e r f o r m a n c e  system i n the  that follow,  results  l i g h t of  expressed  characteristics  i n d i v i d u a l l y ; those considered  favorable  enumerated  are  dealt  first.  5.2.1  Favorable  1)  Characteristics  The f i r s t  objective,  p e r f o r m i n g a c q u i s i t i o n and p a t t e r n time i s  s u f f i c i e n t l y fast 2)  Instead input,  Processing  performed  immediately  proceeding process  i n this  recognition  is  before 3)  inherently  until  feature after  need  entire  extraction  the  an a d d e d  b a s i s was  strokes  the  final  intermediate  point  is  or  also  r'esponse  adopted.  c h a r a c t e r s have  is obtained. discarded early  storage. after  The f a c t  been are  By in  the  that  a c q u i r i n g the  (and o c c a s i o n a l l y t h e  a final  decision  is  sampled).  The c l a s s i f i c a t i o n scheme  insensitive  system  By  and p a r t i a l c l a s s i f i c a t i o n  pressure vector  advantage  fulfilled.  recognition.  redundant data are for  is  concurrently,  d e c i s i o n may b e made v e r y s h o r t l y  pressure value rendered  operation,  on a p o i n t - b y - p o i n t  fashion,  o b v i a t i n g the  analysis  to a l l o w o n - l i n e  of d e l a y i n g a n a l y s i s pre-processing,  real-time  employed i n the  to s i z e v a r i a t i o n s  author's  system  since positional information  not  analyzed.  This independence  is  almost without  are  moderately  f a m i l i a r w i t h the  system perform  is is  q u a l i f i c a t i o n ; u s e r s who  indifferently with  characters  102 of  1mm a n d l m .  Examples of reasonably  Appendix B where 4)  a l l characters  Since spurious  were  extreme  symbol s i z e s  end-point  data  fourth objective i s  are  treated  extraction,  the  reduces  the  computing burden s i n c e the q u a n t i t y  subsequent  s t a g e s i s l o w e r e d and t h e  A l a r g e p r o p o r t i o n o f the pre-processing 5) advantageous The a u t h o r any  research  in  correctly recognized.  feature  overall  appear  also  before  fulfilled.  This  after  approach  o f data passed  complexity of l a t e r  effort  and n o t  stages i s  to  reduced.  was expended i n o p t i m i z i n g  the  stage. B e r n s t e i n and H o w e l l c l a i m e d t h a t  to  the  adopted  design of a  approach  p r a c t i c a l r e c o g n i t i o n system  s u c h an a p p r o a c h ;  d e c i s i o n s made,  a pragmatic  [BERN69],  each d e s i g n s t e p r e s o l v e d t o  i m p l e m e n t i n g , f i n e - t u n i n g and then  was  testing  justifying  the  final  version. 6) character entirely gross  Miller's  slant  and p o s i t i o n [ M I L L 6 9 ] .  independent  distortions  quantization  and f a i r l y  scheme w i t h  rotations  16° h y s t e r s i s  tolerance  to  very tolerant  at  character  to o v e r a l l  owing to  segment  the Xebec pen  the  slant,  course  borders.  is  direction  A technique  rotation i s outlined  for  in  5.5. 7)  (eg.  severe  i n s e n s i t i v i t y to  Performance w i t h  o f p o s i t i o n and i s  p r o v i d i n g even g r e a t e r Section  s c h e m e e x h i b i t e d some p e r f o r m a n c e  Although Ledeen's  system suffered  s e v e r a l major  a l p h a n u m e r i c s U and V w e r e i n d i s t i n g u i s h a b l e ) , t h e  implementing with [LEDE73].  l o w CPU a n d l o w memory r e q u i r e m e n t s  The a u t h o r has  Memory r e q u i r e m e n t s  are  restricted  minimal  microcomputer compatible.  his  feasibility  was w e l l  system i n a s i m i l a r  (discussed  later)  and t h e  Not a l l p r e v i o u s l y r e p o r t e d  drawbacks of  established  fashion.  CPU b u r d e n  systems  is  operate  in  modest e n v i r o n m e n t s .  Groner, for example, u t i l i z e s  i m p l e m e n t an 8 - t u p l e to be s u f f i c i e n t 8)  extraction process;  f o r the Xebec pen-based  In Chui's  intentionally tolerant  feature  thesis  S y s t e m 360  a single feature  was  to found  system.  are s e v e r a l pages o f s u c c e s s f u l l y r e c o g n i z e d  distorted characters  to e i t h e r  an I B M  [CHUI76].  The a u t h o r ' s  system i s no  less  l o c a l o r g r o s s v a r i a t i o n s as e v i d e n c e d b y e x a m p l e s i n  Appendix B . 9)  Chui's  minimal user  training.  the pen a f t e r tablet  scheme e x h i b i t e d a s e c o n d f a v o r a b l e c h a r a c t e r i s t i c ,  a few m i n u t e s o f p r a c t i c e .  systems where  and r e q u i r e s requires  He r e p o r t e d t h a t u s e r s h a d n o d i f f i c u l t y  the o p e r a t o r  a minimum o f  although i t  Fortunately  the s k i l l  10)  seems t o b e e a s i l y to be  The a u t h o r ' s  increased  t o a minimum. 1 3k  Rand  Rom  5 1 2 RAM Serial interface  1  Display  scheme  training. i.e.  scheme i s  CPU a n d memory r e q u i r e m e n t s h a v e  8 - b i t CPU  1  tablet  the  repeated.  A complete system would comprise the  device  (LED, LCD, e t c . )  S e v e r a l TTL gates  for  approximately  r e t a i n e d by o p e r a t o r s ,  As w i t h C r a n e ' s system, the Xebec pen-based  microcomputer c o m p a t i b l e [CRAN77]. kept  case  Performance s t a b i l i z e s a f t e r  improves s l i g h t l y w i t h  l e a r n i n g p e r i o d s h o u l d n o t have  the  m u s t v i e w a CRT w h i l e w r i t i n g o n a  an h o u r o f t r a i n i n g .  some f a m i l i a r i z a t i o n .  15 m i n u t e s  This i s not  operating  following:  entirely been  104 All  10 f a v o r a b l e c h a r a c t e r i s t i c s  incorporated into  t h e p r o p o s e d scheme.  were t o be e l i m i n a t e d from the  5.2.2  Elimination  1) purpose enter  of  The a u t h o r  ambiguous  (the  pair is  if  time-out  was d e s i g n e d  discussed  which  next.  are q u i t e  t r i e d to  character  unambiguious c h a r a c t e r  set  avoid  pairs,  d i s s i m i l a r and t h u s  this  for  e.g.  the  subjects  easily  r e s t r i c t i o n by i n c l u d i n g  I>yv0', ' - W ^ ' a n d "  the second s t r o k e  certain  "  of =  to extensive  facilities  c o n s i d e r e d t o be a s y s t e m l i m i t a t i o n .  may b e  the  " neglected  It  is  the  user.  desirable  systems. which  Xebec pen-based  are  the  scheme  reduce  user  acceptance  a single constraint  that  only  analysis  does  author's  G r o n e r ' s scheme i m p o s e d f a i r l y  could  scheme  minimal.  t o p r o v i d e an e n v i r o n m e n t  This idea helped motivate  computing  By e x t r a c t i n g  and s i m p l i f y i n g  r e q u i r e d computing f a c i l i t i e s  3)  personalized  (a 4 - d i r e c t i o n i n d i c a t o r )  a heavy  The a u t h o r ' s  to operate i n a s m a l l machine environment.  calculations,  design of  severe  o f the  stylistic  system.  was s t i p u l a t e d :  not  In  the  an o e r s t e d  would  zero. 4)  performed  In B e r n s t e i n ' s  on two l e v e l s  as p r i m i t i v e s t r o k e s Dividing  12 c h a r a c t e r i s t i c s  from d i g i t s z e r o t o n i n e he had  ambiguous s i n c e  one t y p e o f f e a t u r e  replace  Aside  S i n c e he h a d a c c e s s  b u r d e n was not  constraints  been  occurs).  2)  constrain  a fairly  C , S , T , X and Z w h i c h  potentially latter  systems have  of Undesirable C h a r a c t e r i s t i c s  system t e s t i n g .  recognizable.  The o t h e r  d e s i g n are  Hosking selected  letters  of other  characters  recognition system,  [BERN68] .  then  Unknown i n p u t s w e r e  relations  into strokes  c l a s s i f i c a t i o n was  between  seemed  the  first  identified  strokes  were  calculated.  t o be an u n n e c e s s a r y  step  and  is  105  therefore  not  contained i n the proposed  5)  T h e h i g h memory r e q u i r e m e n t  [BERN69] was j u d g e d system i s for  an u n d e s i r a b l e  o f B e r n s t e i n and H o w e l l ' s characteristic  t o o p e r a t e i n a s m a l l machine environment.  t o t a l software  computer  t o be  system.  ( i n c l u d i n g d i c t i o n a r y ) the  author's  s i n c e the  A t under  scheme  proposed  1500  system i s  words  micro-  compatible. l " 6)  character  Miller  limited  s i z e i s not  character  constant  was j u d g e d u n f a v o r a b l e .  size  t o -j-  [MILL69] .  i n a p r a c t i c a l environment,  Since  handprinted  such a  The a u t h o r ' s s y s t e m i s i n s e n s i t i v e  to  constraint  character  size. 7) complete.  Miller  In the  concurrently, and a t  the  character  [LEDE73] .  especially  recognition, [CHUI76].  in  the  repertoire  Chili,  performed  CPU t i m e d u r i n g d a t a  input  (i.e.  users had to e n t e r s e v e r a l copies before  classification decisions  this  users  restrictions a l l the  more  of  each  could  be  undesirable  environment.  as  constrains  p l a c e d on a h a r d s u r f a c e .  proposed system,  environment  are  a q u i s i t i o n was  recognition decision delay  a pen and t a b l e t  Unfortunately,  Furthermore,  of free  l i k e many r e s e a r c h e r s i n t h e  selected  of paper  use  data  T h i s method o f s y s t e m d i c t i o n a r y d e f i n i t i o n , i s  i n a "many u s e r " 9)  until  response).  I n L e d e e n ' s scheme,  member o f t h e  analysis  a u t h o r ' s s y s t e m t h e s e two p r o c e d u r e s  same t i m e r e d u c i n g t h e  8)  sheet  to postpone  thus making e f f i c i e n t  improving system  made  chose  are  are not  free  input  field  character  device for his  system  t h e w r i t i n g medium t o  a  Since a tablet  required  t o use  i s not  single  a variety of suitable  p l a c e d on c h a r a c t e r  acceptable.  of  s i z e making  surfaces. the  106 10)  Although the  acceptable  the  attempting  to a v o i d t h i s  quantity  quality of his  analysis  shortcoming,  conclusions  drawn from e x p e r i m e n t a l  sample s i z e  allowed  11) the  number  to  classification  but neglected  report  reduce  effort  sections  In  was c a l c u l a t e d i n c a s e s  often  fairly  where  complicated,  s h o u l d u s u a l l y be m i n i m i z e d . performs  of data  important  w a s made t o  many s i m p l e  that i s  Unlike  pre-processing  u l t i m a t e l y passed  to  the  system  experimental r e s u l t s .  In  conduct meaningful experiments  description  this  thesis,  and to  fully  results. a l l 10 f a v o u r a b l e  characteristics  competitive  characteristics  were i n c l u d e d and  have been e l i m i n a t e d from the  system i s not,  a number o f p r o b l e m s a r e  however, without  discussed  proposed  fault.  and a b r i e f  In  comparison  the is  systems.  System L i m i t a t i o n s  The i n h e r e n t to  s i g n i f i c a n c e of  statistical  results  author  amount  The X e b e c p e n - b a s e d  made w i t h  5.3  the  to quantify  12 u n d e s i r a b l e  next  the  C r a n e a n d S a v o i e p u b l i s h e d an i n t e r e s t i n g  subsequent  scheme.  references  [CRAN77] t h e  I n summary, all  improved.  stage.  12)  substantial  c o u l d have been  Since d i c t i o n a r y searches are  Crane and S a v o i e  quite  this.  of necessary  calculations  o f t e s t s p e r f o r m e d by C h u i i s  unavoidable  groups.  (diagonal)  zero)  o f the  feature  extraction  d i f f i c u l t i e s i n d i s c r i m i n a t i n g amongst  F o r example  technical  simplicity  the  i s not  stroke  request  that  z e r o be  m e r e l y an a e s t h e t i c  drawn as  factor.  stage  certain  i t w o u l d be i m p o s s i b l e t o d i s t i n g u i s h i t  lead  character  an o e r s t e d  Without the  can  (or  second  from a s i x .  107 In F i g . 5 . 1 ( a ) ,  d i r e c t i o n codes  and d i g i t s i x a r e  indicated.  f o r the  The n e c e s s i t y  A s i m i l a r problem could occur Illustrations with  a loop  i n Figure 5.1(b)  (which i s not  the  ANSI n i n e  variations the  o f some  the  characters  the  built  the  loop  is  s i x and  is  cannot  be  tolerated  i n t o the  since  they  proposed  Non-technical users  remains  that  rate  the  For example, L i n  approach  t h e n been,  to  been  local upon  the  modify the selected,  defined  features  automatically This  computing  added requirements.  Systems  d i f f i c u l t i e s the Xebec pen-based  system  schemes.  and r e s p o n s e t i m e .  expense of a lengthy  problem  encroach  can n e i t h e r  to extract  by increased  c r i t e r i a f o r j u d g i n g the  recognition  six.  system concerns  when n e c e s s a r y .  Comparison w i t h P u b l i s h e d  Two m a j o r  the  c o u l d have  however,  competitive with existing  other.  dictionary  been w r i t t e n  which  Despite potential  nine  pen-  a serious  obvious, however,  l i m i t a t i o n o f the  d i c t i o n a r y and r e f e r e n c e d  Performance  a  others.  p r o g r a m c o u l d have  unknown c h a r a c t e r ,  nine.  for  and n o t  t h i s was n o t  zero  apparent.  for a non-nominal  It  H a d an e x a c t - m a t c h  c o n v e n i e n c e w o u l d be o f f s e t ,  at  oersted  d i r e c t i o n code  set.  configuration.  repertoire.  a learning-type  5.4  digit pair  technical  r e c o g n i t i o n l o g i c f o r e x i s t i n g s y m b o l s n o r a d d new s y m b o l s  system's  of  exhibit  standard  A second p o s s i b l e dictionary  the  may b e m i s t a k e n  d i r e c t i o n code d e s c r i p t i o n o f  chosen  the  always v i s i b l e s i n c e p r e s s u r e s  does n o t  when t e s t i n g w i t h t h e  for  the  for  i n d i c a t e how t h e  motion generate d i r e c t i o n codes) Since  familiar zero,  reports 4 second  success  I n many c a s e s ,  o f any s y s t e m i n c r e a s i n g one  a high recognition score response  time  [LIN74] .  of  are decreases 99.4%  Groner's  but  D i f f e r e n t i a t i n g the Zero, Oersted, and S i x  F a m i l i a r Zero LDRULQ (Nominal)  Oersted  Six  LDRULODQ  LDRULQ  (Nominal)  D i f f e r e n t i a t i n g the S i x and Nine  F i g u r e 5.1  (Nominal)  109 scheme,  on t h e  half-hour  responds  t r a i n i n g s e s s i o n were not  performance and h i s  other hand,  [GRON66].  Also  the  " q u i c k l y " but  tested  able to obtain b e t t e r than  latter  utilized  a n a l y s i s / c l a s s i f i c a t i o n system alone  3 2 - b i t words of  subjects  addition, i t  r e q u i r e d more t h a n  system requires  natural  responds  the  almost  and p o t e n t i a l l y l e s s  system i s  expensive  response  t h a n most o t h e r  The a u t h o r ' s  an o v e r a l l  s y s t e m compares  u s e r b e c o m e s p r o g r e s s i v e l y more f a i m i l i a r Chui's which i s  even s m a l l e r than  linearity effects paper,  Computek p e n and t a b l e t  etc.)  writing.  the  tablet  96% t h a t  a t w r i t i n g medium e x t r e m e s .  r e s t r i c t i o n severely limits  appears to  increases  low  as  the  the  to a w r i t i n g  system suffers  The w r i t i n g s u r f a c e  the  be  r e c o g n i t i o n accuracy  users  c a n n o t b e p o s i t i o n e d more t h a n ^ " f r o m t h e  This  systems.  operation.  constrains since  less.  more  favorably with i t s  w i t h pen  area  an h o u r o r  schemes  character  t i m e and a r e c o g n i t i o n a c c u r a c y above  levels,well  space-consuming,  b e t t e r s m a l l machine environment  t h a t p r o p o s e d b y C h u i who r e p o r t s 98.3% [Chui76].  less  entire  dictionary) .  immediately w i t h performance  Xebec pen-based  One o f t h e  of  2400  o n l y 1500 1 6 - b i t words f o r t h e  90% f o r i n e x p e r i e n c e d u s e r s h a v i n g p r a c t i c e d f o r h a l f  Furthermore,  an 87%  a l a r g e IBM System/360  p r o g r a m ( i n c l u d i n g a q u i s i t i o n , a l l s t a g e s o f a n a l y s i s and the  above  a  code.  The a u t h o r ' s  In  after  tablet  area  non-  (pad o f  s u r f a c e when  t y p e o f h a r d - c o p y medium  allowed. The c o m p u t e k p e n i t s e l f pressure  must be a p p l i e d b e f o r e  processor. often  In Chui's  o m i t t e d due  thesis,  is  awkward and h e a v y .  p o s i t i o n a l data  he noted  is  A 3-ounce  communicated to  t h a t p a r t i a l or e n t i r e  to i n s u f f i c i e n t p r e s s u r e .  This  is  vertical the  strokes  e s p e c i a l l y true  host were  for  light writers. controllable  Under the  and i s  author's  i n c l u d e d as  system  the  v e r t i c a l threshold  is  a parameter i n the p e r s o n a l i z e d  software'  scheme.  3 Chui  stated  that  characters less  than  t a l l were  o  drawn w i t h  3 difficulty. formatting  Groner,  in his  implementation,  s q u a r e s and c l a i m e d t h a t  used  t h e s e were  coding forms w i t h — "  of a f a i r l y  typical  high  size.  1 The X e b e c pen p e r f o r m s  well with less  than - j ^ " h i g h  Chui chose to u t i l i z e s e v e r a l agreed w i t h t h i s direction position rapid  information [POWE73].  "front  obtained w i t h the  5.5  i n certain  superior  to  Recommendations Before  might be  familiarization,  stood  large  above  user,  s h a l l be  (rather than  natural  to h i m .  the  system  feature,  successive,  character  adequate  direction.  recognition  a u t h o r does n o t i n this  Neither  since  pen  for  systems  know o f a n y t h a t  is  thesis.  Research  especially  a number d u r i n g the  system philosophy o f hardware early  stages  that  modifications of  examined.  hands found  extreme s i t u a t i o n  the  improve performance,  the  relative  u s i n g one p a r a m e t e r  authors'  extracted  one p r o p o s e d  Hand-size varies cularly  i n the  of handprinted  Further  as  cases of low d i s c r i m i n a t i o n .  single  areas,  using  c h a r a c t e r s and l e a v i n g  d i s c u s s i o n changes i n r e c o g n i t i o n  a p p l i e d to  that would a s s i s t  the for  for  implemented  A l t h o u g h a number  generally  o f unambiguous  techniques  these s u g g e s t i o n s were  appear to e x c e l  Powers  i n c o n j u n c t i o n w i t h a n o t h e r parameter such  A l t e r n a t i v e l y , Powers a d v i s e d  complex,  r e s u l t s were  c l a s s i f i c a t i o n methods.  o f m u l t i - l e v e l r e c o g n i t i o n and s u g g e s t e d  end" d e c i s i o n s  p e r h a p s more of  concept  characters.  but pen b a r r e l  shape i s  t h a t when h o l d i n g the i n contact  user  functions  with)  fixed.  Users with  pen c o m f o r t a b l y ,  the w r i t i n g s u r f a c e .  i n an e n v i r o n m e n t  A s i m p l e s o l u t i o n w o u l d be  to p r o v i d e  that for  the In  i s not  partipen-tip  this completely  adjustment  of  I l l the  barrel/pen-tip  angle.  A n o t h e r method o f c o n s t r a i n i n g u s e r s t o h o l d v e r t i c a l p o s i t i o n w o u l d be t o barrel) adopt  near  the  the pen b a s e .  configure  a traingular  Alternatively,  l i k e device that would attach  just  one m i g h t p r e f e r  above t h e p e n - t i p  m o t i o n ' w h i l e p r e v e n t i n g t i l t i n g movements.  natural  with  than  placement circular  a c o n v e n t i o n a l pen/paper suggestion  any a s s i s t a n c e  posture  forming each  before  During experiments the  subjects'  the  vertical position»  The p r e s e n t  worthwhile increased.  i s not  5 " than — i n diameter 16  are  array.  the  Damped  commonly a v a i l a b l e .  by v e r i f y i n g  conducted the  pen  a u t h o r p l a y e d an a c t i v e p a r t  b y a d v i s i n g when t h e  pen had d e v i a t e d role i f  in  from a  developed. adequately  under  B u t what w o u l d be the  and o n s y s t e m r e s p o n s e completely?  user  character.  system functions set.  in  less  calls for  The b u b b l e - l e v e l c o u l d p l a y a s i m i l a r  symbol s p e c i a l character  basket-  environment.  the novice could b e n e f i t  l e a r n i n g process  c o m m e r i c a l s y s t e m were t o be  or redefined  to  w o u l d be c o n s t a n t l y m o n i t o r e d a n d a l t h o u g h an e x p e r i e n c e d  would not need  performance  author  f o r e n s u r i n g good p e n p o s t u r e  l e v e l s measuring l e s s  the  an i n v e r t e d  s y s t e m would t h e n be  o f a s m a l l b u b b l e - l e v e l above t h e s t r a i n - g u a g e  Pen a t t i t u d e  (of  a l l o w i n g up a n d down p e n  However, the  u s i n g such a d e v i c e s i n c e the  A final  cross-section  I n t h i s way t h e p e n w o u l d be more l i k e l y  desired posture.  agreement  the pen i n a n o m i n a l l y  time i f the  I n any e x t e n s i o n  to examine t r a d e - o f f s  set  the  effect  defined  on r e c o g n i t i o n  were e x t e n d e d  further  to t h i s w o r k , i t w o u l d  that would a r i s e  19-  as r e p e r t o i r e  be  size  is  112 Constructing personalized dictionaries/systems into  the  effect  of various  factors  conclude that  extreme of  cases,  these factors  drag pressure  optimized independently c l e a r how t h e y  affect  w o u l d be u s e f u l  Chapter I I I ; digits  minimum l e f t  stroke  strokes  their aid  concept  user  rate)  trials,  each  can it  be is  not  system.  It  were e x p l a i n e d in.  is  Consider the  applied,  r e q u i r e d to produce  Without h y s t e r e s i s ,  average  While  i n t o the  examined h e r e .  an  two  inferred  an " L " i n  the  any d i r e c t i o n e x c e e d i n g 45° w o u l d  p o i n t may s e e m i n s i g n i f i c a n t a f i n i t e  are p a r t i c u l a r l y s u s c e p t i b l e .  or hinder the  subject  is  w i l l be m i s r e c o g n i z e d because o f i t .  strokes  testing  of hysteresis  hysteresis  e x c e e d i n g 53° i s  Although this  levels.  in  inter-dependences.  disadvantage  When t h e  vector.  be s u f f i c i e n t .  these  possible  several did,  and sample  when r e - i n t r o d u c e d  and e f f e c t s  a potential  direction-code  of  one a n o t h e r  for  i n F i g . 5.2.  threshold  i n conducting r e p e t i t i v e  to examine  The r e a s o n s  robust,  some c o n t r o l u p o n p e r f o r m a n c e  (e.g.,  By  o f one a n o t h e r . . i t was  a l t h o u g h most p a r a m e t e r s were  exercise  insight  upon r e c o g n i t i o n p e r f o r m a n c e .  a s s u m i n g s y s t e m p a r a m e t e r s t o be i n d e p e n d e n t to  provided  number  W r i t e r s who c a r e l e s s l y  To what  extent  can o n l y be a s c e r t a i n e d  hysteresis  through  form effects  further  testing.  G = 55 : A c c e p t a b l e w i t h or without hysteresis. F i g u r e 5.2  0 = 53 : A c c e p t a b l e without but u n a c c e p t a b l e with h y s t e r e s i s .  Negative E f f e c t s of H y s t e r e s i s  113 Participants if  in recognition trails  one was t o draw a c h a r a c t e r  system could rotations (refer the  accomodate  to Appendix B ) .  quadrant  plane  code v e c t o r w i t h or decremented)  each  I n the  elemental  changed s i n c e  as  rendered  Tables VIII  the  likely  to have  performance. suffer  if  It  this  aware o f p e n  c o u l d a l s o be t o l e r a t e d  event  direction rotated It  remains  pattern  analysis  greater  rotating  the  directionincremented  techniques  unknown, however,  effect  t o what  were  decisions  are  results).  i n d e f i n i n g the  (and p o s i t i v e )  f r e e d o m was r e m o v e d  by  ( i . e . , modulo-4  and I X . f o r q u a n t i t i a t i v e  a pronounced  that  the  w o u l d o n l y be w o r t h w h i l e  i f the c u r r e n t  d e s i g n e r has  that  mechanics  o f an " U n k n o w n " i n p u t  s y s t e m s t a n d s v e r y few " U n k n o w n " i n p u t  The f r e e d o m t h e is  u s e r was w e l l  c o u l d be r e t e s t e d .  i m p l e m e n t i n g s u c h a scheme  (cf.  i f the  happen  They w e r e t o l d  o f up t o a p p r o x i m a t e l y 4 5 ° a n d  But rotations  itself.  asked what w o u l d  90° or 180° rotated.  rotations  c o u l d be t o l e r a t e d  often  character  upon  recognition  extent  (i.e.,  i f symbols were  thesis  are  repertoire  the  system  would  t o t a l l y defined  a  priori). Results obtained worthwhile  to determine  accommodated last  i n this  whether  using only inferred  remark i s  information [POWE73].  at  one's  noted e a r l i e r  sets could  directions  for their  description.  o f the that  case o f p e r s o n a l i z e d  t h a t h a v i n g b o t h p o s i t i o n a l and  T h i s may r e q u i r e  properties  character  d i s p o s a l would improve performance  X e b e c pen w o u l d be l o s t . the  even l a r g e r  e s p e c i a l l y true i n the  Powers suggested  e n c o u r a g i n g and i t w o u l d be  This  systems. directional  considerably  d e v i c e and a major a d v a n t a g e s o f  A preferable  force  to the  a tablet  a p p r o a c h m i g h t be  i n f o r m a t i o n generated  at  the  be  to  further  pen-tip.  a u t h o r ' s knowledge w r i t i n g v e l o c i t y was  the  examine It  was  fairly  114 unpredictable even w i t h i n  differing  a given symbol.  a s o l u t i o n to In it  f r o m day t o d a y ,  the the  shorter-lived  case o f the  than  be d i f f e r e n t i a t e d assumption  is  the  area might u l t i m a t e l y  non-significant  define of  assume t h a t  " L " i n the  " l "stroke  the  of the  constant,  s i x and n i n e o f F i g . . 5 . 1 ( b ) , loop of the  s i x and t h a t  the  nine  is  two d i g i t s  could  The u n d e r l y i n g  short-lived implies small could  lead  t h e way  classification.  an a b s o l u t e  successive  this  i.e.  i f prove'd g e n e r a l l y t r u e ,  l o o p s and l o c a l  Again,  provide  direction information. digits  of character  lengths  nominal strokes.  s y m b o l t o s y m b o l and  by d i c a r d i n g a l l s h o r t - l i v e d d i r e c t i o n s .  Rather than  that  final  to user,  ambiguous p a i r ,  This assumption,  level  c o u l d compare  to  spurious  t h a t pen v e l o c i t y i s  displacement. a secondary  Research i n t h i s  problem of  i s probably reasonable  user  (or  direction duration suspect)  variations  i d e a would have  threshold,  directions are  one  hypothesizing  of shorter  duration  t o be s u b s t a n t i a t e d  by  than  further  experimentation. Inter-character considered i n the proposed laboratory user  One o f t h e  extent  of a s t r o k e .  strokes  applicable available.  strokes are  simplest  to  implement  space,  it  i n a non-  "prompts" would d e t r a c t have been  involves determining  By d e f i n i n g a s t r o k e ' s l i e in this  for  A number o f schemes  f r o m a common c h a r a c t e r .  to the  not  i m p l e m e n t a t i o n b u t w o u l d be i m p o r t a n t  from h i s p r i m a r y i n t e r e s t .  successive  o r d e l i m i t a t i o n was  a p p l i c a t i o n since having to wait  past.  not  segmentation  p h y s i c a l space  applied i n  the  i s not  whether  presently  a u t h o r ' s system since p o s i t i o n a l i n f o r m a t i o n i s  the  geometrical  and v e r i f y i n g  i s possible to estimate T h i s scheme  the  not  that or  to  115 A second p o s s i b i l i t y strokes; that  i.e.,  concerns  the t i m e e l a p s e d between  a lengthy i n t e r v a l suggests a completed symbol.  0 . 3 s e c o n d s was o p t i m a l f o r e x p e r i e n c e d u s e r s  technique,  u s e r s are  forced to  reduce  [GRON66].  successive Groner found  Using  this  t h e i r normal h a n d w r i t i n g speed  by  a p p r o x i m a t e l y 50%. A more e l e g a n t subsequent If  not,  stroke  the  could  a p p r o a c h w o u l d be t o reasonably  s t r o k e must be t h e  regarding the  determine whether  be an e l e m e n t  first  of the previous  of the subsequent  p r e v i o u s symbol c o u l d be communicated to the  neither  effectiveness  constrained to  [CRAN76]. fixed  nor to reducing his writing  i n using this  character speed  sizes  user.  technique  (as i n geometric  ( d a t a may b e e n t e r e d  the  Research to determine whether  is  user  time and core  are  r e c o m m e n d a t i o n one c o u l d c o n s i d e r t h e  t i m e - s h a r i n g p e r s o n a l i z e d systems  t o answer q u e s t i o n s  for  several users.  at  i n workbooks at  a main c o n s o l e .  their  Similarly  desks  and  the  at  additional  possibility  arises.  personalized  Again,  f o r each o f  s h o u l d be  the o p e r a t o r s .  undertaken.  the  and t e a c h e r s m o n i t o r  e a c h d e s k and t h e  A study o f the  pens their  application i n a  ( n u m e r i c a l or o t h e r )  a p e n c o u l d be i n s t a l l e d at  of  S u c h an i m p l e m e n t a t i o n  the pen might f i n d  commerical environment where u s e r s e n t e r d a t a  required  bursts  required  m i g h t b e u s e f u l i n an e d u c a t i o n a l e n v i r o n m e n t w h e r e s t u d e n t s u s e  need  is  j u s t i f i e d w o u l d be w o r t h w h i l e . As a f i n a l  progress  remark  delimitation)  i n short  Additional  effort  do n o t  the  commands c a n be c o n c a t e n a t e d ) . dictionary creation stage.  formation.  s y m b o l and a d e c i s i o n  C r a n e and S a v o i e h a v e i m p l e m e n t e d s u c h a scheme b u t as t o i t s  the  as  the  subsystem  c o m p u t i n g power  116 5.5  Conclusions  A simple, effective, operating  i n a limited  accurately  and  Results to  characterize  in  this  for  the  indicate  are  entire  that  generally  time i s and  c o m p i l e d r e c o g n i t i o n scheme  accepted  to e f f i c i e n t  and e l i m i n a t e s  the  the  however,  need  restricted  level  o f 95% a r e  techniques.  for  inherent  for  automating  implemented  successfully.  the  The A N S I h a n d p r i n t e d tested  of s e l e c t i n g  minimal, core  bit-words  dictionary. performance  Recognition accuracies  exceeding  e a s i l y a c h i e v e d and s y s t e m  response  a c q u i s i t i o n and a n a l y s i s  The p e n - p a p e r  the  may b e a t t a i n e d  robustness tolerance  generation  numerics  i n a simulated applications  importance  fairly  Furthermore,  i n e x p e n s i v e and i t s  user-dedicating)  system because of the scheme's  is  used  environment  stages  is  natural  retranscription.  (i.e.,  A technique  the  including  concurrency of the  comparable performance  because of the  were  small,  sufficient  a p p r o x i m a t e l y .1500 16  system improves  by r e d u c i n g imposed s t y l i s t i c c o n s t r a i n t s .  that  reasonable  gesture repertoire  is obviated.  e x i s t i n g schemes.  analysis  Personalizing acceptance  any s i z e and any  defined numerics plus  microcomputer c o m p a t i b l e at  l o w due t o  scheme  to  information alone i s  computing f a c i l i t i e s  very competitive with  the  symbols o f almost  drag-force  The p r o p o s e d s y s t e m i s is  and i m p l e m e n t e d  S i n c e the number o f c a l c u l a t i o n s r e q u i r e d  for extensive  requirements  real-time  distortion.  symbols o f the  thesis.  and n a t u r a l  e n v i r o n m e n t was d e s i g n e d  recognize handprinted  degree o f s l a n t  the need  core-efficent  familiar  It  i s worth  under  a less  o f optimal parameter to undefined  noting, natural,  values  character  of p e r s o n a l i z e d systems  set  operator  variations. was  and n i n e s p e c i a l g e s t u r e  environment.  symbols w h i c h are  Results software  and  symbols  demonstrated defined  in  117 t h e p r o p o s e d s y s t e m and t h e r e f o r e  easily  The r e c o g n i t i o n scheme i s w e l l p e r f o r m at  least  special writing  as w e l l surfaces  c a t i o n performance w i t h component i n f u t u r e  modified. r e c e i v e d b y n o n - t e c h n i c a l u s e r s who  as t e c h n i c a l u s e r s . nor extensive  facilities  low system response  system's  design.  Such a system r e q u i r i n g and o f f e r i n g  time c o u l d prove a  neither  good c l a s s i f i useful  118 REFERENCES [ANSI74]  American National Standard Character Set for Handprinting (ANSI 43.45), ANSI, Inc., New York,  [BATC74]  B. Batchelor, P r a c t i c a l Approach to Pattern C l a s s i f i c a t i o n , Plenum Press, London, New York,  [BERN68]  1974.  1974  M. J . Bernstein, "A Method for Recognizing Handprinted  Characters  i n Real-Time", i n Pattern Recognition, L.N. Kanal, Ed., Thompson Book Co., Washington, [BERN69]  1968.  M.J. Bernstein and H. Howell, "Handprinted  Input for On-Line  Systems", Rep. NASA-CR-12B4, Systems Development Corp., Santa Monica, C a l i f o r n i a , [CASE64]  1969.  R. Casey and G. Nagy, "Recognition of Printed Chinese Characters", IEEE Trans, on E l e c t . Computers, Vol. C-15, December  [CCI 78]  No.l, pp. 750-752,  1964.  "Experimental Signature V e r i f i c a t i o n System Detects Forgeries", Canadian Controls & Instruments, p. 5, November  [CHUI76]  T.K.  1978.  Chui, "Real-Time Computer Recognition of Handprinted  Characters", M.A.Sc. Thesis, The University of B r i t i s h Columbia, 1976.  [CLAR73]  R. Clare, Designing Logic Systems Using State Machines, New  [CRAN77]  H.D.  York,  1973.  Crane and R.E.  Handprinted  McGraw-Hill,  Savoie, "An On-Line Data Entry System f o r  Characters", Computer Magazine, pp. 43-50, March  1977.  119  [DEV067]  D . B . Devoe, Data", pp.  [DIM057]  "Alternatives  IEEE T r a n s ,  21-32,  March  to H a n d p r i n t i n g i n t h e  o f Human F a c t o r s  i n Electronics, Vol.8,  Eastern  of  No.  1,  1967.  T . L . Dimond, "Devices f o r Reading H a n d - W r i t t e n Proc.  Manual Entry  Joint  Computer C o n f e r e n c e ,  pp.  Characters",  232-237,  December  1957.  [DUDA68]  R . O . Duda, P . E . H a r t , Handprinted pp.  [FRAN78]  [FREE61]  1139-1149,  -  Recognition of Proc.  FJCC,  1968.  1978.  506-510,  i n the  Context A n a l y s i s " ,  pp.  - Applied Informatics,  H . Freeman,  "On t h e  IEEE T r a n s ,  on E l e c t r o n i c Computers,  Encoding of A r b i t r a r y Geometric  Vol.11,  Configurations",  V o l . EC-10, pp.  260-268,  1961.  H . Genchi et  pp.  [GRON66]  II  Augewandte I n f o r m a t i k  Characters  [GREA63]  Part  H.W. Franke,  June  [GENC68]  Text:  "Experiments  al.,  "Recognition of Hand-written  for Automatic L e t t e r  1292-1301,  E . C . Greanias  August  et  al.,  pp.  1963.  G . F . Groner, AFIPS P r o c .  Vol.56,  "The R e c o g n i t i o n o f H a n d - W r i t t e n N u m e r a l s  IBM J .  January  IEEE,  1968.  Contour A n a l y s i s " , 14-21,  Sorting", Proc.  Numerical  of Res.  and D e v . , V o l . 7 , N o .  "Real-Time Recognition of Handprinted FJCC , V o l . 29, p p .  591-601,  1966.  1,  Text",  by  [GRON68]  G.F.  Groner, "Real-Time R e c o g n i t i o n of Handprinted  P a t t e r n R e c o g n i t i o n , L.N. Washington,  [HOEL62]  [HOSK72]  K a n a l , Ed., Thompson Book  143-145, John W i l e y  K.H.  Hosking,  Co.,  1968.  P. H o e l , I n t r o d u c t i o n to Mathematical pp.  Symbols" i n  & Sons, New  Statistics,  York,  3rd  Edition,  1962.  "A Contour Method f o r the R e c o g n i t i o n  o f Handp r i n t e d  C h a r a c t e r s " , i n Machine P e r c e p t i o n of P a t t e r n s and P i c t u r e s ,  The  W h i t e f r i a r s P r e s s , L t d , , London, 1972 .  [ICHI741  T. Ichikawa and J . Y o s h i d a ,  "On-Line R e c o g n i t i o n of  Handprinted  C h a r a c t e r s w i t h A s s o c i a t i v e Read-Out of P a t t e r n s i n a Memory", P r o c . of Second I n t e r n a t i o n a l J o i n t Conference R e c o g n i t i o n , pp.  [ITO  78]  M.R.  I t o and T.K.  Standard Vol.  [LANS70]  [LEDE73]  1974.  C h u i , "On-Line Computer R e c o g n i t i o n o f Proposed  ANSI (USASI) H a n d p r i n t e d  10, pp.  341-349,  Characters", Pattern  Recognition,  1978.  J . Lansdown, "Computer A r t f o r T h e a t r i c a l Performance", i n Proc,  [LANS78]  206-207, August  on P a t t e r n  ACM  I n t e r n a t i o n a l Computer Symposium, Bonn,  J . Lansdown, "The  Computer i n Choreography", Computer Magazine,  Vol.  11, No.  K.S.  Ledeen, see,"The Ledeen C h a r a c t e r R e c o g n i z e r " ,  of  8, pp.  19-31, August  I n t e r a c t i v e G r a p h i c s , W.M.  McGraw-Hill,  1979.  I n c . , New  York,  1978.  Newman and R.F. 1973.  i n Principles  Sproull,  121  [LEVI69]  [ L I N 74]  M.D. Levine,  "Feature C l a s s i f i c a t i o n :  Vol.  1391-1405,  57,  pp.  W . C . L i n and T . L . S c u l l y ,  [P0WE70]  G. M i l l e r ,  FJCC, V o l . 35, pp.  G. Nagy, " S t a t e  of  Vol.  pp.  56,  No. 5,  V . M . Powers,  [TEIT64]  the  W. S t a l l i n g s ,  836-862,  497-504,  399-412,  1969.  Proc.  IEEE,  May 1 9 6 8 .  Ph.D. dissertation,  from  University of  Sequences i n C h a r a c t e r  V o l . 5, pp.  "Probability  291-302,  and S t a t i s t i c s ,  Recognition,  December  1973.  McGraw H i l l  Book C o . ,  1975.  "Computer  Characters",  C Y . Suen,  Proc.  pp.  Trans,  Symbols",  Art i n Pattern Recognition",  "Pen-Direction  M.R. Spiegel,  Description  AFIPS P r o c .  and R e c o g n i t i o n  FJCC, V o l . 44, pp.  " H a n d w r i t i n g Models i n Canadian  of  Printed  1015-1025,  Education",  from C o n c o r d i a U n i v e r s i t y , M o n t r e a l , March  W. T e i t ,  IEEE  1970,  V . M . Powers,  Report  Constrained  " O n - L i n e R e c o g n i t i o n o f Hand-Drawn C h a r a c t e r s  San F r a n c i s c o ,  [SUEN79]  of  V o l . SMC-4, N o . 6,  "On-Line R e c o g n i t i o n of Hand-Generated  Proc.  Chinese  IEEE,  1974.  Pattern Recognition,  [STAL75]  Identification  M a n , and C y b e r n e t i c s ,  Michigan,  [SPIE75]  "Computer  on S y s t e m s ,  Directional Information",  [POWE73]  of  1969.  Characters with a High Recognition Rate",  AFIPS  [NAGY68]  Proc.  Handprinted  November  [MILL69]  August  A Survey",  Interim  1979.  " R e a l - T i m e R e c o g n i t i o n of Hand-Drawn C h a r a c t e r s " ,  FJCC, V o l . 26,  part  I,  pp.  559-575,  1964.  1975.  AFIPS  .  122  [TOU 72]  J.T.  Tou and R . C . G o n z a l e s ,  "Recognition of Hand-Written  by T o p o l o g i c a l F e a t u r e E x t r a c t i o n IEEE T r a n s ,  [TRUA78]  B . Truax, System",  on C o m p u t e r s ,  and M u l t i l e v e l  V o l . C-21, pp.  "Computer M u s i c C o m p o s i t i o n : Computer M a g a z i n e ,  Characters  Categorization",  776-785,  July  1972.  The P o l y p h o n i c POD  V o l . 1 1 , N o . 8,  pp.  19-31,  August  1978.  [WONN72]  T . H . and R . J . W o n n a c o t t ,  Introductory S t a t i s t i c s ,  2nd E d i t i o n , John W i l e y & S o n s ,  [WOOD76]  J . Wood, 6 t h University  Congress  of  the  Toronto,  p. 2 2 1 ,  1972.  I n t e r n a t i o n a l Ergonomics  of M d . , College Park,  J u l y 11-16,  1976.  Association,  APPENDIX A Example o f R e c o g n i t i o n System  Performance  ( D i g i t s c i r c l e d were i n c o r r e c t l y classified, o v e r a l l performance exceeds 98%)  t~3  7  1 r  5"  C  V ?  U7  4 ~  ^  °J  ^  ^  ^9-  C5  ^ A x /  -  VO  1 ^H ^ s>  1  124  APPENDIX B Example of D i s t o r t e d Character Forms  APPENDIX C  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0094674/manifest

Comment

Related Items