一种基于奇异值分解和置信传播的图像匹配算法

An Image Matching Algorithm Based On Singular Value Decomposition and Belief Propagation

图像匹配技术是计算机视觉中一个很重要的问题。当匹配在不同的视角、不同光照、局部遮挡以及复杂背景的情况时,由于特征的可重复性以及区分性下降会导致许多误匹配。针对上述问题,提出了一种提高图像匹配精度方法,这种方法能够去除误匹配的同时恢复丢失的匹配点对。首先采用快速鲁棒性特征(Speeded Up Robust Features,SURF)提取关键点和描述子,从而构建邻接矩阵,然后对邻接矩阵进行奇异值分解获得初始匹配。三角剖分用于提纯初始的匹配,最后通过双图限制恢复丢失的匹配点对。在Oxford数据集测试的实验结果表明,所提出的方法在匹配性能和精度方面,优于随机抽样一致性算法(Random Sample Consensus,RANSAC)。与此同时,该算法的稳定性也相应提高。

Finding correspondences between pair of images of the same scene is a key problem in computer vision. When matching images undergoes viewpoint change, partial occlusion, clutters and illumination change, there will be a lot of mismatches due to the limited repeatability and discriminative power of features. A robust matching algorithm that can remove false matches and propagate the correct ones to obtain more matches is proposed, thus improve the matching accuracy. Firstly, SURF (Speeded Up Robust Features) descriptors of each image are extracted, which can be used to build the proximity matrix. Secondly, SVD (Singular Value Decomposition) is performed on the proximity matrix to obtain the initial matches. Thirdly, the unique property of delaunay triangulation is adopted to refine the initial matches which can produce the maximum clique of the two delaunay graph. Finally, the lost matches are recovered with the constraint of dual graph of Voronoi. Experimental results on Oxford datasets indicate that the algorithm can improve match performance compared to the RANSAC-based method. At the same time, the stability of our method is better than RANSAC.