多尺度边缘动态规划立体匹配算法  

Multi-scale stereo matching using edge dynamic programming

作　　者：肖俊[1] 夏林元[1,2] 林丽群[3] 肖洲[4] 

机构地区：[1]武汉大学测绘遥感信息工程国家重点实验室,武汉430079 [2]中山大学地理科学与规划学院,广州510275 [3]湖北大学资源与环境学院,武汉430062 [4]北京市测绘设计研究院,北京100038

出　　处：《测绘科学》2012年第1期142-144,共3页Science of Surveying and Mapping

基　　金：国家自然科学基金(40874007)

摘　　要：目前大多数立体匹配方法都是基于像素处理,而以面向对象的方式进行立体匹配则更加符合人类视觉习惯。该文提出一种面向对象的多尺度递进的立体匹配算法,实现由粗到细的快速立体匹配。首先对参考图进行粗分割获取目标对象,然后针对目标区域采用边缘线动态规划的方法获取视差,再根据精度需求进一步判断该目标是否需要细分以获取更加精细的视差,最后得到有效的视差图。实验证明,这种方法以目标区域为对象避免了从像素级逐一运算,不仅使处理速度加快,而且还能保持有效精度。For the weakness of the pixel-based stereo matching,setting segment as a processing element is more practical.This paper presented a stereo matching method using an object-oriented multi-scale method to achieve from coarse to fine stereo matching.Firstly,from the perspective of human stereo vision,the reference image was segmented by an object-oriented segmentation method with large scale,which simulates human perceptual process and fully takes shape,spectral and spatial context into account.Secondly,by employing a refined D Palgorithm which is based on the inner edgeline of the segment,the segment＇s disparity was obtained.Thirdly,according to the accuracy requirements,the segment was determined whether need to be further subdivided to obtain more precise disparity map.Finally,the final disparity image was obtained.The experimental results indicated that consistent with human being＇s vision,this algorithm combining segmentation with refined D Palgorithm sped up stereo matching process while keeping matching results accurate.

关 键 词：立体匹配 图像分割 面向对象 动态规划 多尺度 

分 类 号：P209[天文地球—测绘科学与技术]