Bioinformatics-Motivated Approach to Stereo Matching

Abstract

We propose a framework for stereo matching that exploits the similarities between protein sequence alignment in bioinformatics and image pair correspondence in computer vision. This bioinformatics-motivated approach is based on dynamic programming, which provides versatility and low complexity. In addition, the protein alignment analogy inspired the design of a meaningfulness graph which predicts the validity of stereo matching according to image overlap and pixel similarity. Finally, we present a technique for automatic parameter estimation which makes our system suitable for uncontrolled environment. Experiments conducted on a standard benchmark dataset, image pairs with different resolutions and distorted images validate our approach and support the proposed analogy between computer vision and bioinformatics. © Springer-Verlag Berlin Heidelberg 2013.