Variable block-size disparity estimation in stereo imagery

Abstract

This thesis addresses the problem of developing and implementing in software a variable block size (quadtree splitting) disparity estimation algorithm that is optimized for use in compression of stereo image pairs and studying its performance over a range of rate/distortion values for a variety of images. First the constrained optimization problem is converted to an unconstrained one using the Lagrange multiplier approach. Then by solving the optimization problem using dynamic programming, the optimal variables representing the optimal quadtree structure and the quantizer for each node are determined. The experimental results show the improvements of this method over simple intraframe JPEG coding and over fixed block-size disparity estimation.