Segmentation-Based,Robust and Fast Local Stereo Matching with Medical Application

A segmentation-based two-step stereo matching is proposed in a fast and local perspective to resolve the ambiguity of binocular stereo problem in this paper.Color segmentation is firstly conducted on both stereo pairs,then robust matching costs are constructed with modified Geman-McClure function and segmentation-based variable support two-pass aggregation only in the horizontal and vertical directions which eliminate ambiguity in feature matching,and then initial disparity map is obtained from them.To assign optimal disparity to ambiguous regions(such as occlusion etc.),greedy disparity estimation consists sequentially of four parts: unreliable disparity detection,segmentation-based narrow occlusion handling,epipolar-line-based least square filling and warping-based disparity combination.The experimental results indicate that this technique can eliminate matching ambiguity and obtain piecewise smooth,accurate and dense disparity map effectively.This algorithm is concise,fast and efficient so that it could provide automatic and reliable depth perception for computer aided diagnosis system in clinical medical application diagnosing diseases.