A Stereo Matching Algorithm Based on Gray-Level Segments
基于灰度段的立体匹配算法

Stereo matching has long been one of the central research problems in computer vision. A stereo matching algorithm is proposed based on gray level segments in order to recover dense depth map for the application of view synthesis in IBR (image-based rendering). This algorithm uses gray level segments as elements to be matched, and introduces the principle of DTW (dynamic time warping), which applied for speech recognition, into the matching processes on the segments. Covering more image space than point elements, gray-level segments are less sensitive to noise and photometric variations and easier to be matched, and can reduce the false matching. Moreover, gray-level segments segments require less computational resources than feature-based line segments and conices. Experimental results show that this algorithm is effective and of great value to use.