快速视差范围估计算法及其应用  

Fast Disparity Range Estimation and Its Applications

作　　者：毛苗[1] 吴艳楠[1] 秦开怀[1] 

机构地区：[1]清华大学计算机科学与技术系,北京100084

出　　处：《计算机辅助设计与图形学学报》2015年第1期120-127,共8页Journal of Computer-Aided Design & Computer Graphics

基　　金：教育部博士点基金(20090002110006);北京市自然科学基金(4102027)

摘　　要：视差范围估计在立体匹配中非常重要, 准确的视差范围能提高立体匹配的精度和速度. 为此提出-种基于匹配代价搜索和图像细分的快速视差范围估计算法. 该算法将输入图像均匀分成多个图像块, 采用匹配代价搜索计算每-图像块的视差, 找到视差最大（最小）的图像块, 并利用迭代细分规则将该图像块继续分成更小的子块, 直至得到稳定的最大（最小）视差; 利用匹配代价图对图像块进行可靠性检测, 以解决弱纹理块容易误匹配的问题. 实验结果表明, 文中算法在保持97.3%的平均命中率的同时将立体匹配的平均搜索空间降低了27.7%, 比采用传统算法可以得到更准确的视差范围; 将该算法应用于立体匹配算法中降低了其平均误匹配率, 并将计算时间缩短了20%～45%.Disparity range estimation is very important in stereo matching. An appropriate disparity rangecan increase the precision and speed of stereo matching. This paper proposes a fast disparity range estimationmethod based on matching-cost search and image subdivision. It evenly divides input images into severalsub-blocks, and searches using matching cost to find out which sub-block is having the maximum/minimum disparity. After that, the maximum/minimum sub-block is recursively divided into smallersub-blocks, until all current sub-blocks have the same disparity. To deal with image blocks with week textures,detecting their disparity reliabilities is applied via matching-cost diagram. Experimental results showthat our method can achieve 27.7% reduction rate of search space while preserving 93.7% hit rate on average.Compared to traditional methods, it can get a more accurate disparity range. Moreover, the gained disparityrange reduces the running time of stereo matching by 20%.45% while decreasing the average false-match rate.

关 键 词：视差范围 立体匹配 迭代细分 图像块视差可靠性 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]