Measurement of Shape from Disparity Based on a Computational Model for the Ocular Dominance Columns in Visual System

Abstract

It is still a difficult problem for machine vision systems to obtain range information, while animals can easily perceive the distance around their environment. Thus learning the functions of biological visual system is essential for the implementation of efficient machine vision systems.
According to E. L. Schwartz, the authors constructed a mock architecture of ocular dominance columns in the visual cortex, by implementing a column-like image with interlaced scan lines taken alternately from left and right cameras. The areas which have horizontal disparities can be extracted by bandpass filtering the column-like image, and hence we can formulate disparity estimation as a problem to minimize the output energy of the band pass filter.
This paper describes derivation of a new algorithm for disparity estimation and presents experimental results which demonstrate the ability of the algorithm to recognize three dimensional environmental structures. It is also shown that this algorithm has the same principle as the profile recovery system using a differential identity of binocular stereo, proposed by S. Ando.