异形窗口快速立体匹配算法研究OA北大核心CSTPCD

The traditional stereo matching algorithms have to choose a suitable support window for each pixel in the image,which is time-consuming and has difficulties in determining the shape and size of the window.To address these issues,this paper proposes a fast stereo matching algorithm of special-shaped window.It selects matching windows based on image gray-scale similarity and spatial proximity,and adds a threshold constraint to the pixels in the search window in the target image to reduce the impact of depth discontinuous regions and occluded regions on the results.Then,the matching calculation is conducted and the calculated parallax is considered to be the parallax of all pixels in the matching window,thus reducing the process of parallax calculation by pixel.The performance of algorithm is verified through numerical simulations using four sets of images from the Middlebury dataset.Our results show the proposed algorithm markedly improves the calculation speed when the accuracy of the disparity map is largely consistent.