Stereo Matching Algorithm Based On Color Information

As an important branch of computer vision, Stereo vision can be widely appliedin many kind of situations such as military surveillance, industrial inspection, objectrecognition, robot navigation, virtual reality, amusement, etc. Stereo matching isbecoming a hotspot which takes a large proportion of whole visual process anddetermines the effect of three-dimensional reconstruction directly and the quality ofthe whole visual system. Stereo matching can be classified into two categoriesaccording to different optimization methods. Global stereo matching algorithm hashigher accuracy, but it is computationally expensive and is difficult to meet the need.Local matching algorithm is widely used by many scholars for its easy and simplemethod, however, it has lower accuracy.In order to improve the matching accuracy many scholars and experts makeunremitting efforts, proposed fixed window algorithm, adaptive window stereomatching algorithm, adaptive weighting algorithm and so on. Adaptive weightingalgorithm has been gained more and more extensive attention because of highermatching accuracy.In this paper, we focus on the difficult problems of stereo matching-repeat orlow texture, occlusion and discontinuous areas, propose improvement which makeuse of the image color information. Experiment result shows our algorithm canimprove the accuracy of repeat, low texture, occlusion and discontinuous areas, andis conducive to the three-dimensional reconstruction and overall stereo visionsystem.In this paper, the improved algorithm made mainly in the following aspects:1.A local stereo matching algorithm based on color is proposed on the basis ofthe analysis of local stereo matching, using adjacent pixels of the same color regionhaving similar disparity, grouping based on color, combining with gradient of absolutedifference and absolute intensity difference in a certain proportion to improve thematching cost; disparity adjustment based on color similarity is given using colorsimilarity results to improve initial disparity map, we test the improved algorithm and analysis the experimental results.2.Region growing algorithm based on the color changing constraint is proposedon the base of the analysis of the adaptive weight algorithm, using regionalconnectivity and color similarity to reconstruct weight factor, we compare theoriginal algorithm and the improved algorithm and analyze the performance ofimproved algorithm.3.An improved adaptive weight algorithm based on mean shift and occlusion isproposed on the basis of analysis of the principle of mean shift algorithm. Weimprove weight support aggregation to cope with occlusion problem, and use meanshift to segment the image, we also make effective improvement in disparity searchbased on color segmentation during matching, finally we present a fast matchingalgorithm combined with region growing and occlusion, improve the occlusion regionmatching accuracy to some extent.