一种改进的自适应权重立体匹配算法与校正

Improved Adaptive Support-weight Algorithm and Adjustment

立体匹配是立体视觉中一个重要的环节,针对自适应权重匹配算法中难以兼顾速度和精度的不足,提出了一种基于色彩的局部立体匹配算法。首先,利用区域连通性和颜色相似性改进权重因子,提出基于颜色变化约束的区域生长算法,有效提高了算法的精度和速度;然后基于颜色相似性提出一种视差校正算法,根据颜色相似性的结果对初始视差图进行视差校正,在没有增加额外工作量的前提下进一步提高算法的精度。实验结果表明,此算法能有效提高重复区域、边缘区域、低纹理区域的精度,而且速度较自适应权重算法提高了近20%,与当前主流算法具有可比性。

Stereo matching is one of the important parts in stereo vision. Regarding difficulties of adaptive support-weight algorithm to meet both accu- racy and speed, a novel local stereo matching method based on color is presented in this paper. Firstly, improving weight factor by using region connec- tivity and color similarity, a region growing algorithm based on color constraint is proposed to improve the accuracy and speed of the algorithm effectively. Secondly, a disparity adjustment based on color is presented to further improve algorithm accuracy without adding extra computation cost. Experimental results show this algorithm can effectively improve the accuracy in repeat and depth discontinuities and low-textured regions and get a better stereo matc- hing quality which is comparable with other main stream stereo methods, speeding up nearly twenty percent.