基于二分图最佳匹配的图像检索方法

摘要

根据图论中二分图最佳匹配的相关知识提出一种新的方法：先将示例图像与被检索图像同时分成n个子块，将两幅图像的每一子块分别看成是图的一个顶点，子块间的相似度作为权值，从而将两幅图像之间的空间对应关系的确定，归纳成为求二分图最佳匹配的问题．经实验证明：该方法对于同类物体由几何变换和空间变换所形成的图像，有着较高的检索性能．%This paper proposed a new method based on the related methods of graph theory. The new method defines every part of the images as a vertex, and then calculates the similarity of as the weight value to design a weighted bi-partite graph. So defining the congruent relationship between two images is converted into searching the optimal matching in this bipartite graph. In the new method, both color dimensional distribution and geometric transformation invariance are considered. The simulation results show the effectiveness of the novel method.