Improving Block-Based Disparity Estimation by Considering the Non-uniform Distribution of the Estimation Error

Abstract

A block-based disparity estimator is proposed that considers the non-uniform spatial distribution of the estimation error inside an image block that is mapped into another image plane using projective 2-D transformations. For this purpose, first the error variance distribution inside the image block is analytically derived. The derived error variance distribution shows four eccentric minima. As a consequence, the proposed disparity estimator arranges the image block eccentrically around the picture element (pel) to be evaluated. Thus, a reduction of the estimation error variance by a factor between 1.5 and 2 can be achieved compared to known block-based disparity estimation techniques. Since the error distribution shows four minima, four possible arrangements of the image blocks and hence four independent estimates can be made. A further reduction of the estimation error variance by a factor up to 2.6 can be achieved, when the four estimates are averaged. Additionally, an outlier detection and removal in the set of four estimates enables an increased robustness. The proposed estimator is tested using both, synthetic image pairs of known disparity and real images. The expected error reduction performance of the estimator and its increased robustness are verified.