Open Collections

Visual echo analysis and multi-evidential correlation: non-linear matching & registration of signals and images

University of British Columbia

Many low-level vision tasks - such as measurement of visual motion, stereo disparity estimation, or texture segmentation - can be solved by similar computational or biological mechanisms. The primary aim of this dissertation is to introduce and describe a broadly applicable approach to address a variety of low level computational vision problems. This unified framework, which is named visual echo analysis, is based on the simple observation that many computer vision problems can be viewed as detection and estimation of echo arrival periods in time and space. To this end, the framework uses cepstral techniques, a common and effective non-linear signal processing methods for detecting the presence of echoes and estimating their spatial or temporal arrival period. The thesis introduces computational and performance improvements to the traditional power and differential cepstrum with direct extensions to complex and phase cepstrum. Visual echo analysis (and multi-dimensional cepstrum) is then applied to a number of low-level vision tasks such as: motion estimation, binocular and trinocular stereo disparity, motion-stereo analysis, multi-frame disparity estimation (multi-frame motion, multiple baseline stereo), stationary texture segmentation, boundary symmetry analysis, and detection and estimation of multiple disparities (i.e., motion transparency, reflection, and occluding boundary). The relationship between echo analysis and matching is briefly examined, and a new technique for signal registration - called multi-evidential correlation (MEC) is introduced. MEC provides multiple, and thus verifiable, measurements for individual point disparities. The technique utilizes specific matching kernels - such as cepstrum, phase correlation or Hadamard based disparity measurements methods - to furnish multiple estimates of individual disparities; estimates that can be used to verify one another and/or be combined to establish a robust and accurate measure of signal displacements.

Thesis/Dissertation

application/pdf

2009-03-19

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.

http://hdl.handle.net/2429/6230

Computer Science

University of British Columbia

UBCV

DSpace