Stereo Matching Based on Dissimilar Intensity Support

This paper presents a local stereo matching algorithm based on the window construction method using local edge detection. In order to improve performance of window-based cost aggregation computation, a new rule called Dissimilar Intensity Support technique is proposed to distinguish support pixels with dissimilar intensities from those with similar intensity for each centered pixel. According to the proposed rule, support pixels away from centered pixel with dissimilar intensities are used to compute cost aggregation for enhancing intensity variation in local support region. Initial disparity is obtained by classical Winner-Take-All method. Then disparity refinement based on pixels near center with similar intensity as well as smooth disparity is introduced to rectify mismatch. The experimental results based on Middlebury stereo benchmark show that the proposed algorithm has good performance especially in low textured region with only few parameters.