基于区域分割的动态规划立体匹配算法

Dynamic Programming Stereo Matching Algorithm Based on Region Segmentation

提出一种基于区域分割的动态规划立体匹配算法。首先参考图像经自适应多阈值切割之后,得到一个由区域组成的集合,并沿着各个闭合区域的边界进行动态规划跟踪,然后对于非匹配区域和区域内部分别作视差融合和视差插值处理,获得最终的稠密视差图。实验结果表明,该算法能够取得较为理想的效果,视差图横向"条纹"瑕疵和边界区域上的误匹配点明显减少,层次更加分明,整个视差图平滑性较好,匹配效率有了显著提高。

A dynamic programming stereo matching algorithm was proposed based on region segmentation. Firstly, the reference image was divided into several different regions by use of adaptive multi-threshold image segmentation. Then, the multi-seed dynamic programming approach was presented for acquiring disparity distribution along the border of closed areas. Finally, parallax in the mismatching area was fused together and interpolated within its area in order to obtain a dense disparity map. The experiment results reveal that the algorithm is better effective both in decreasing the overall matching error rates on the boundary of the domain and reducing apparent defective stripes. Moreover, its efficiency of matching is improved significantly.