均衡化概率模型及其在特征匹配中的应用

Balanced probabilistic model and its application on feature matching

针对计算机视觉研究中最基本的特征匹配问题,提出了一种新的特征匹配算法,通过对匹配的概率模型中邻接矩阵的均衡化分析完成匹配过程。该算法根据待匹配特征点的几何关联性,建立匹配的概率模型,并利用一种有效的双向均衡方法,使得待匹配特征点的关联权重达到平衡,从而增强了匹配的区分度和准确性。采用重启动的随机游走(randomwalkswith restart,RWR)方法对概率模型进行求解,结合具有约束限制的时序方法获取最优匹配集。实验结果表明,与同类方法相比,该算法具有更高的匹配准确率,且适用于多种匹配场合。

Aiming at the feature matching,the basic problem of computer vision,a new algorithm of feature matching is proposed by balancing analysis of adjacency matrix of the matching model in a probabilistic framework.The approach according to all the interaction of the two candidate feature point sets and then a probabilistic model is established,then using an efficient method for bidirectional,which make all the relevance weight balancing,and it improves the discriminative and accuracy performance of matching.Finally,the probabilistic model is solved using random walks with restart（RWR）,and correct matches are recovered by imposing a sequential method with mapping constraints in a simple way.Compared with the similar algorithm,the method is accurate in terms of matching rate in various matching applications.