TOP SIFT Based Image Retrieval

Multimedia information has been growing fast with the rapid growth of Internet in recent years,such as video,audio,images,etc.How to search an object image quickly,accurately and effectively from large dataset has been a difficulty and a focus of multimedia application research.With the rapid growth of unlabeled images in the internet,and the highly requirement of image retrieval,the text-based image retrieval is not sufficient to the needs of the people on the current image retrieval applications,the researchers gradually move their focus on content-based image retrieval,since content-based image retrieval is more in line with the needs of people in the real application,it has received extensive attention and been focused on multimedia application research..SIFT is a local descriptor that proposed by David Lowe in 1999,and modified by him for further improving and perfecting in 2004.SIFT is a local image feature that could description local interested points in an image,and contains rich of information of image.SIFT feature descriptor is invariant to uniform scaling,orientation,and partially invariant to affine distortion and illumination changes.Based on the above characteristics,various SIFT based feature extract algorithms has been proposed and has been successfully used for a variety of computer vision applications,SIFT based image retrieval is one of the algorithms in content-based image retrieval.However,SIFT algorithm is time consumption,high complexity,and cost a lot of memory when the image datasets is very large,since it can generate much features in an image.In addition,it probably contains redundant information in the set of SIFT features that may cause error matching in image retrieval.Many researchers proposed methods to improve SIFT algorithm to solve above problems,most of them apply dimension reduction approach to improve the retrieval efficiency while not reduce the retrieval accuracy.These methods have achieved good effect,but still not solve the problem of it may exist redundancy information in the features.How to find more respectively SIFT points is still an urgent research problem.In view of the above problems,this paper focus on SIFT based image retrieval.The main contents of this paper are as follows:1.In this paper,we first make a lot of research on image retrieval,then review the theory of global image feature and local image feature in details,and analysis of their respective advantages and disadvantages.2.In this paper,we make research on the background of SIFT algorithm,and deeply study the theory in details,study the SIFT application in content-based image retrieval,and points out that the shortcomings and improvement direction of SIFT in image retrieval.3.In this paper,we propose an novel SFIT selection algorithm based on PageRank that could selected the most respectively SIFT points in a set of SIFT features,we called TOP-SIFT,it significantly improved the speed of image retrieval.4.In this paper,we present a novel method to calculate the similarity between images,Multi-Matched.The experiments show that the combination of Multi-Matched and TOP-SIFT significantly improved the speed of image retrieval and retrieval accurate.