Local Stereo Matching Based on Information Entropy of Image

Abstract

Adaptive support-window algorithm is one of the simplest local algorithms for stereo matching. An important problem for adaptive support-window algorithm is to determine the appropriate support-window size, which is always hard to do and limits the validity of adaptive support-window algorithm. An appropriate support-window size must be selected adaptively based on image features. In this paper, information entropy of image is defined for stereo matching in the RGB vector space. Based on adaptive support-window, a new support-window selection algorithm, which uses information entropy of image to quantify image features such as illumination color and number of object contained in an image, is proposed. Experimental results evaluated on the Middlebury stereo benchmark show that our algorithm outperforms the conventional adaptive support-window algorithms.