Multiple Feature Association Image Content Retrieval Algorithm And Its Application Research

With the development of the multimedia and the wide application ofthe computer， the image resources grow rapidly, how to extract the usefulvisual information of images has became the key research, which makesthe image retrieval has became a very active field of research.1970s, thetechnology of image retrieval was mainly based on the text, but themanual annotation with strong subjectivity and incompleteness.Content-based image retrieval is a very good solution to this problem, themain idea is that extract visual features of image directly (such as color,texture, shape), the user submits features of image,the system matches thefeatures of image in the database of image automatically and displays theimage of close similarity to the user. Use the feature of image for imageretrieval is the core of the method.Based on the basis of relevant information widely at home and abroad,research to the technology of Content-based image retrieval. Most of theexisting integrated features found in image retrieval algorithm fails toreflect the spatial distribution of information of contents in image fully,the results of retrieval is just passable. For the problem of this kind of search,this paper propose a retrieval algorithm which based on area of interest ofimage with multiple and correlative feature,denoised the image first,segmented the image which denoised first into several regions, then,extracted points of interest of image with Gabor filters, thereby determinethe area which follow with interest closely by the user. Be extractedfeatures of image respectively from the area of interest which followclosely by the user based on color histogram, texture feature based on graylevel co-occurrence matrix, shape feature based on moment invariants,matched similarity feature of image with external normalization finally.Research of the retrieval method is applied in image databasesystem, the experimental results show that the algorithm not only makeslower dimension with the extracted feature vectors, but also ensures theinvariance of rotation, translation and scale, it has higher retrievalefficiency.