Mean Shift的渐进无偏变换图像配准.pdfVIP

第 34 卷第 2 期 电 子 与 信 息 学 报 Vol.34No.2 2012 年 2 月 Journal of Electronics Information Technology Feb. 2012 Mean Shift 的渐进无偏变换图像配准 杨 烜* (深圳大学计算机与软件学院 深圳 518060) 摘 要：基于标志点对应关系的图像配准方法不能保证变换的无偏性，一般不适用于大形变配准问题。该文提出了 一种基于标志点对应关系的渐进式无偏变换图像配准方法，该方法利用少量的初始标志点对，通过 Mean Shift 迭 代寻找对应标志点对，构造无偏性较好的变换函数，逐步得到一致的配准结果。文中详细讨论了标志点的分布对变 换函数无偏性的影响，实验结果表明该文方法可以同时适用于大形变和小形变配准，是一种简单、有效的配准方法。 关键词：图像配准；无偏图像变换；对称Kullback-Leibler (KL)距离；拓扑保持 中图分类号：TP391 文献标识码： A 文章编号：1009-5896(2012)02-0393-05 DOI: 10.3724/SP.J.1146.2011.00689 Progressive Unbiased Image Registration Using Mean Shift Yang Xuan (College of Computer Science and Software Engineering, Shenzhen University, Shenzhen 518060, China) Abstract: The landmark-based registration methods do not define an unbiased transformation between the two images, which are not suitable for registering images where there are large deformations. A progressive landmark-based image registration method using Mean Shift is presented. This method is capable of providing unbiased transformations and defining a consistent correspondence between two images. Based on few initial landmark pairs, new corresponding landmark pairs are searched by Mean Shift and tacked in the images gradually to construct unbiased transformations. The progressive registration is conducive to Mean Shift to search corresponding points using satisfied initial position, and avoid Mean Shift dropping into the local minima. Experiments show that the method is simple and feasible to large deformation and small deformation registration problem in the meantime.