基于全局拓扑结构的分级三角剖分图像拼接

Global Topology Based Image Stitching Using Hierarchical Triangulation

采用相似性度量的方法对具有周期性内容或相似内容的图像进行配准时,容易产生特征误匹配,从而带来拼接误差.针对这一问题,提出基于全局拓扑结构的分级三角剖分图像拼接方法:首先,提出基于梯度及3色比空间的特征描述用于相似性度量,保留所有阈值范围内的m:n(m,n为正整数)特征点匹配,以减少漏匹配;然后,根据特征点集的拓扑结构对特征点集进行分级三角剖分,根据三角形网格的匹配关系,逐步将多对多的不确定匹配或降为一对一匹配,去除误匹配.实验结果表明,与经典图像拼接方法相比,该方法可以解决周期性内容或相似内容误匹配带来的拼接误差,并大大减少投影变换矩阵计算点数.

Most image stitching algorithms adopt intensity or gradient for similarity measurement.Unfortunately they fail when the scene exhibits periodic contents or similar contents.In this puper,we propose a novel global topology based image stitching method.Firstly,gradient and ratios of RBG are used to describe and compare feature points.In order to reduce the omission of matched points,a threshold is set to reserve all m：n（m,n are positive integer） feature matching.Then,we compute two compatible triangulations of gradient and color matched points to compare the topology similarity.Finally,we incrementally add 1：1 matching points which are topology matched and remove problematic points.We demonstrate the illustrative results by comparing and contrasting our output with other methods.The presented algorithm gives superior results in all examples.