基于引导滤波和视差校准的自适应权重算法

PDF全文下载： 2016020227

肖静文a， 万敏b

 (西南石油大学 a.电气信息学院; b.机电工程学院，四川 成都 610500)

 摘要：  提出了一种基于导向滤波器和视差校准的自适应权重算法。首先，使用梯度与色彩相结合的相似测度计算不同视差的原始匹配代价，构成三维代价卷，使用导向滤波器对其进行滤波，有效保持边缘信息，提高了初始匹配精度,并通过积分图像方法降低了运算复杂度；然后，增加了视差校准步骤，根据色彩相似性和空间相近性分配自适应权重，对初始视差图进行校准，进一步提升匹配精度。利用Middlebury网站提供的标准立体图像对进行匹配精度验证，结果表明：算法可以有效消除歧义匹配，获得精度较高的稠密视差图。通过真实图像对算法进行了验证，能够在一定程度上恢复景物的三维信息。

 关键词：  双目视觉；立体匹配；自适应权重；引导滤波；视差校准

 中图分类号： TP 391.41         文献标志码： A

 Adaptive Supportweight Algorithm Based on Guided Filtering and Disparity Calibration

 XIAO Jingwena, WAN Minb

 (a.School of Electrical Engineering and Information； b.School of Mechatronics Engineering,

SouthwestPetroleumUniversity,Chengdu610500,China)

 Abstract: An adaptive supportweight algorithm based on guided filtering and disparity calibration is raised by this article. Firstly, the similarity measure of combination of gradient and color is used to calculate the initial matching costs of different disparity in order to construct the three dimensional cost volume. Through the filtering of guided filter, the edge information is preserved effectively and the initial matching accuracy is improved greatly. In addition, the computational complexity is reduced by the integral image method. Then, the disparity calibration procedure which distributes adaptive weight according to the color similarity and spatial proximity is added to calibrate initial disparity map, thus further improves the matching accuracy. Standard stereo image pairs provided by Middlebury website were utilized in the experiment. The results demonstrate that the algorithm can eliminate ambiguity matching effectively and obtain high accuracy dense disparity map. The algorithm which has the ability of recovering the threedimensional scene information to some extent is verified through the real images.

 Key words: binocular vision; stereo matching; adaptive weight; guided filtering; disparity calibration

 收稿日期： 20150104

基金项目： 四川省科技计划项目（15ZC0923）.

作者简介： 肖静文（1988—），女，硕士研究生.

 文章编号： 16726987（2016）02022706； DOI： 10.16351/j.16726987.2016.02.021