Study On Stereo Matching Based On Phase Information

Stereo matching is the key content in stereo vision, which makes it possible to reproduce a 3-D stereo by getting the distance information through pixels. It has been applied successfully in more fields, such as feature detecting, object recognition, medical image processing. Phase-based matching method is of high accuracy, good robustness and parallel computing structure. And it is the main research direction of stereo matching.Upon analyzing the existed stereo algorithms thoroughly, this paper proposes a new phase-based matching approach, to solve the problem of phase wraparound and singularities existing in current phase-based methods. And a new kind of matching measure function is constructed. Given a particular disparity, the stereo task can be translated into finding certain disparity which can implement best matching between ideal phase differences and measured ones. It integrates the information from each scale filter, so it can effectively achieve high accuracy and wide disparity range.In addition, to satisfy the real-time request, upon analyzing the phase differences and the instantaneous frequency method, this paper also proposes a fast phase-based matching algorithm. The algorithm provides direct evaluations of phase differences of image pairs from the even and the odd of Gabor filters. It can avoid explicit calculations of single phase, obtain high computing efficiency, and is suitable for software and hardware implementing.