关键词: 分割一致性检验, 立体匹配, 超像素分割, 视差精化

Abstract:

Aiming that the accuracy problems in texture lacking images when using the traditional approach, an improved method based on entropy rate super-pixel segmentation is proposed. The Mean-shift algorithm is replaced by the entropy rate super-pixel segmentation algorithm. Firstly, the reference image is segmented into super pixels. Then, statistical analysis is carried out for each segment so that the trusted and untrusted values can be screened according to the central tendency value. Finally, the disparity map is obtained by holes filling. Fifteen groups of images in Middlebury datasets are selected for disparity acquisition and detection. The experimental result shows that, this improved method has a good optimization effect for both texture lacking images and complex texture images, the average mismatching rate of this algorithm is 5.88 percentage points lower than that of the traditional algorithm.

Key words: segmented consistency check, stereo matching, super-pixel segmentation, disparity map refinement