| With inputs of slightly different projections from the two eyes, stereopsis provides a sensation of depth, and perception of a vivid three-dimensional world instantaneously without ambiguity. Creating computer algorithms that can automatically accomplish stereopsis is an interesting yet challenging task that has drawn considerable attention in the areas of computer vision, psychology, and physiology for decades.;In this dissertation, we employ new modulation (AM-FM) models for image representation, a multichannel (Gabor wavelet) processing paradigm for multi-scale (coarse-to-fine) computation of local image phase, and a disparity channel model for depth computation. These new techniques are used in the development for a new stereo algorithm that generates a dense, sub-pixel accuracy disparity map without sophisticated feature extraction and interpolation. We also describe recent psychophysical experiments that suggest that the human visual system uses a similar coarse-to-fine strategy; comparisons between the psychophysical data and the results of our computational model are shown.;Range information computed from stereopsis in machine vision is significant for many practical applications. These include manufacturing inspection, autonomous robotics, automated cartography, and tracking of moving targets among many others. |
