An Improved Image Segmentation Based Stereo Matching Algorithm

In order to decrease the mismatching problem in low texture and occluded areas, an improved algorithm based on image segmentation is proposed. Firstly, the adaptive polygon-based support window is adopted for initial disparity estimation, then the reliable matching points derived through left-right constraint are used to compute the disparity planes in the segmentation areas; finally the disparity planes are used to constraint the disparity estimation and as a reference of the energy function, then the Tree DP is introduced to compute the optimal disparity for every pixel. In order to testify the validity and practicability of the algorithm, firstly we use it in the matching of the standard images, the experimental results show that our algorithm has a better performance in areas of low texture and occluded areas and a higher matching accuracy, and then the reconstruction results of the real images shows a smooth and valid point cloud model.