Local stereo matching algorithm: Using small-color census and sparse adaptive support weight

Abstract

This paper proposed an effective disparity estimation algorithm based on census transform with adaptive support weight, called small-color census and sparse adaptive support weight (SCCADSW). Census transform provides high resistance to radiometric distortion, vignette, and noise because it are based on the relative ordering of local pixel intensity values rather than the pixel values themselves. This transform is widely used in many computer vision applications. A simplification technique such as using small-color census is used to determine the initial matching cost. The color distances are transformed using small census transform to keep the information of the color. To derive support weights, Manhattan distances are used for all pixels of the support window to the window's center point. Property of adaptive support weight leads to improved segmentation results and consequently to improved disparity maps. This work is still on process, to test the algorithm; it will use the Middlebury benchmark. According to analysis of each step of the algorithms, the proposed SCCADSW can achieve good performance among stereo methods that rely on local optimization.