An efficient disparity algorithm based on adaptive window selection for 3D reconstruction in stereo

Abstract

In this paper a novel matching algorithm is proposed that improve the quality of disparity map in order to increase the accuracy in stereo 3D reconstruction algorithm. In the stereo matching algorithm, moving windows over disparity map is used. However choosing an unsuitable window size can affect the quality of the disparity map. For instance, by using a large window size, the details may be lost; in contrast a small window can detect details while produces much noise. In this paper we try to benefit from the advantages of these large and small window sizes by computing disparity map using adaptive window size. For this reason, at first a disparity map with small window size is computed. After that, noise in this disparity map is detected, and replaced by the same pixel in the produced disparity map base on a large window size. Our simulations show that this procedure results in a high quality map with less noise and more details.