Local Stereo Matching with Edge-Based Cost Aggregation and Occlusion Handling

This paper presents a new local binocular stereo matching algorithm. A self-adapting matching score is used to measure the dissimilarity of two corresponding pixels. In matching cost aggregation step, we replace the cost of each pixel with the average cost of selected neighboring pixels based on edge pixels information to limit cost aggregation within the same segment, which can preserve the shape and size of the discontinuity boundary. In disparity refinement step, an occlusion model is set up and occluded pixels are detected based on the model to improve the disparity maps accuracy. Experimental results using the Middlebury stereo test bed demonstrate the good performance of our proposed approach.