Stereo matching based on support points propagation
Hua Wu, Zhan Ping Song, Jian Yao, Liang Li, Yu Gu · 2012
In this paper, we present a novel propagation-based stereo matching algorithm. Firstly, we select highly reliable depth points (i.e., support points) from both the high-textured areas and the low-textured ones in the rectified images. Then we propagate these support points along adjacent neighboring structure to produce the depths of other points in the image based on the triangulation derived from a set of highly reliable support points. This allows for efficient exploitation of the disparity search space, yielding accurate dense disparity without the need for global optimization. Experimental results successfully demonstrate both the effectiveness and matching accuracy of our proposed method.