摘要
提出了一种以K-均值分割为基础的立体匹配方法.该方法不仅可以根据图像的内容自动调整匹配窗口的形状,还可实现对参与匹配窗口的大小、数目和权重的智能调节.作者采用K-均值分割方法精确定位物体边界,保证匹配窗口位于同一物体内部;邻域限制与放松可以进一步根据图像内容灵活地运用匹配窗口周围的环境信息;两种方法的结合有效地提高了匹配过程中窗口选取的智能性.在国际立体视觉标准平台Middlebury网站中测试的结果证实该算法提取的深度图的错误率低于其它局部优化算法,接近全局优化算法,运行效率高于现有的全局优化算法,综合性能是出众的.
The paper proposes a stereo matching method based on K-means Segmentation.Within the proposed method,more than one matching windows are involved in one corresponding task.Not only the shape but also the size,number and weight of each matching window can be modified intelligently according to the image content.K-means Segmentation was used to detect object edges and kept the matching window within a same object.Neighborhood Constraint and Relaxation Algorithm is further adopted to utilize the environment information.This combination tackles the problem of how to choose an appropriate matching window intelligently.The algorithm is tested using the Middlebury stereo test bed.It was proved that the error percentage of the depth map obtained by the algorithm is lower than other algorithms based on local optimization,approaching to global optimization algorithms.The efficiency of the algorithm is higher than other global optimization algorithms,the overall performance is outstanding.
出处
《计算机学报》
EI
CSCD
北大核心
2011年第4期755-760,共6页
Chinese Journal of Computers
基金
北京市重点学科建设项目(XK100080537)资助
关键词
立体匹配
邻域限制与放松
K-均值分割
邻域权重设置
遮挡处理
stereo matching
neighborhood constraint relaxation
K-means segmentation
neighborhood weight setting
occlusion handling
