Study Of Image Similarity Based On Mixed Graph Structure

In this thesis,we first introduce some classic image similarity algorithms, and discuss the advantages and disadvantages of them.Then, we summarize common definitions of image similarity.On these theoretical basis, we propose a new image similarity algorithm based on the mix-graph structure.The Attributed mix-graph is a special way of image description,it’s a dynamic structure.Region nodes can be merged according to the principles of connectivity、similarity and comparability,this decreases the computation of the algorithm and improve the efficiency.Traditional image similarity theories are mostly about the whole part comparison between images,but the algorithm in this paper is different.We can extract the similar parts of images.The reduction of mix-graph structure consider the region integrity of the images and the two different thresholds can ensure the reliability of the algorithm. After these two steps,we can get the similarity value of the similar parts between images.By experimental analysis,we can prove that this algorithm has achieved the initial success, It greatly improve the similarity of two images and extract the effective similar contents between images.