- Library Home /
- Search Collections /
- Open Collections /
- Browse Collections /
- UBC Theses and Dissertations /
- Iterative depth from defocus (I-DFD) algorithms
Open Collections
UBC Theses and Dissertations
UBC Theses and Dissertations
Iterative depth from defocus (I-DFD) algorithms Sewlochan, Ray
Abstract
Depth From Defocus (DFD) seeks to obtain depth information from camera images by measuring the amount of defocus in the images. Although DFD avoids many of the problems associated with stereo vision by maintaining only one point of view, other issues such as optical modelling and computational intensity determine speed and accuracy of DFD systems. This thesis presents the optical theory that establishes the relationship between defocus and depth. Defocus is modelled by the convolution of a scene and a defocus kernel to produce a defocussed image. Dependencies on the scene are removed by measuring the relative defocus between two images of the scene. This is modeled by the convolution of the sharper image with a relative defocus kernel to produce the blurrier image. The determination of the relative defocus kernel is the challenge of DFD. Spatial domain algorithms avoid the errors and assumptions associated with frequency domain transformations. Iterative techniques provide an accurate way of finding the relative defocus kernel, at the expense of computational intensity required by convolution calculations. This thesis develops a hierarchy of Iterative DFD (I-DFD) algorithms for the spatial domain. The hierarchy is divided into defocus modelling method (theoretical (T) or measured (M)), relative defocus shaping method (Gaussian (G) or regularized (R)) and convolution implementation (two-dimensional (2), separable (S) or integrated (I)). Four of the algorithms (TR2, TRI, TGS, TGI) in this hierarchy are implemented and fully tested on three classes of images. TR2 has been previously published and is used as a benchmark. TGS and TGI significantly reduce computational effort, but suffer from degraded accuracy. TRI reduces computational effort by several orders of magnitude with no degradation of accuracy for two of the three classes of images. Recommendations are made to alleviate the degradation in the third class of images.
Item Metadata
Title |
Iterative depth from defocus (I-DFD) algorithms
|
Creator | |
Publisher |
University of British Columbia
|
Date Issued |
1995
|
Description |
Depth From Defocus (DFD) seeks to obtain depth information from camera images by
measuring the amount of defocus in the images. Although DFD avoids many of the problems
associated with stereo vision by maintaining only one point of view, other issues such as optical
modelling and computational intensity determine speed and accuracy of DFD systems.
This thesis presents the optical theory that establishes the relationship between defocus and
depth. Defocus is modelled by the convolution of a scene and a defocus kernel to produce a
defocussed image. Dependencies on the scene are removed by measuring the relative defocus
between two images of the scene. This is modeled by the convolution of the sharper image with
a relative defocus kernel to produce the blurrier image.
The determination of the relative defocus kernel is the challenge of DFD. Spatial domain
algorithms avoid the errors and assumptions associated with frequency domain transformations.
Iterative techniques provide an accurate way of finding the relative defocus kernel, at the expense
of computational intensity required by convolution calculations. This thesis develops a hierarchy
of Iterative DFD (I-DFD) algorithms for the spatial domain.
The hierarchy is divided into defocus modelling method (theoretical (T) or measured (M)),
relative defocus shaping method (Gaussian (G) or regularized (R)) and convolution implementation
(two-dimensional (2), separable (S) or integrated (I)). Four of the algorithms (TR2, TRI, TGS,
TGI) in this hierarchy are implemented and fully tested on three classes of images. TR2 has been
previously published and is used as a benchmark.
TGS and TGI significantly reduce computational effort, but suffer from degraded accuracy.
TRI reduces computational effort by several orders of magnitude with no degradation of accuracy
for two of the three classes of images. Recommendations are made to alleviate the degradation
in the third class of images.
|
Extent |
6983716 bytes
|
Genre | |
Type | |
File Format |
application/pdf
|
Language |
eng
|
Date Available |
2009-01-16
|
Provider |
Vancouver : University of British Columbia Library
|
Rights |
For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use.
|
DOI |
10.14288/1.0064877
|
URI | |
Degree | |
Program | |
Affiliation | |
Degree Grantor |
University of British Columbia
|
Graduation Date |
1995-05
|
Campus | |
Scholarly Level |
Graduate
|
Aggregated Source Repository |
DSpace
|
Item Media
Item Citations and Data
Rights
For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use.