平衡图匹配方法在多视角图像配准中的研究  

ON REGISTRATION OF MULTI-VIEWING ANGLE IMAGES BASED ON BALANCED GRAPH MATCHING

作　　者：刘智[1] 胡福乔[1] 赵宇明[1] 

机构地区：[1]上海交通大学自动化系系统控制与信息处理教育部重点实验室,上海200240

出　　处：《计算机应用与软件》2014年第8期201-204,264,共5页Computer Applications and Software

基　　金：国家自然科学基金项目(61175009)

摘　　要：在多视角图像处理中,多点匹配是一个基本的问题。针对有较多制约因素的多点匹配问题,对现有的光谱匹配技术进行两点改进:首先提出一种新的光谱松弛技术用以求解匹配函数的近似解,其次将现有的相容性矩阵重构为二分图边矩阵并进行双随机归一化,减小噪声对结果的影响。在基于CAVIAR与PETS2009数据集上的实验证明,所提出的算法可以很好地完成图匹配,并且有较高的精度与较好的鲁棒性。Graph matching is a fundamental issue in multi-viewing angle image processing. Two improvements are made on existing spectrum matching technique in the paper aiming at multi-point matching issue with quite a few constraint factors： The one is the new spectral relaxation technique for the approximate solutions to the matching function. The other is to reconstruct current compatibility matrix to a bipartite graph edge matrix followed by bistochastic normalisation that reduces the influence of noise on the effect. The experiment based on CAVIAR and PETS2009 datasets proves that algorithm presented in the paper can implement the graph matching very well,and has high accuracy and good robustness.

关 键 词：图匹配 双随机归一化 多视角 光谱匹配 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]