Superpixel and Plane Information Based Disparity Estimation Algorithm in Stereo Matching

In stereoscopic image processing, disparity estimation helps determine the depth or distance information of objects. In this paper, a more accurate algorithm is presented to estimate the disparity map based on data processing techniques. Different from most of the existing methods, which mainly apply the features around the key points, we apply the plane information extracted from superpixels to perform disparity estimation. Since the key points, such as the scaled invariant feature transform (SIFT) points, are distributed nonuniformly and many regions may not contain any key point, using only the information of key points may not work well in some cases. This problem is well addressed by the proposed superpixel-based plane information. After carrying out entropy rate superpixel (ERS) segmentation, we perform disparity estimation for each superpixel based on the point, edge, and plane information of the current and surrounding superpixels. Experiments show that the proposed algorithm outperforms other methods and can much improve the accuracy of disparity matching.