Research Of Relevance Feedback Image Retrieval Based Clustering

With the rapid development of modern information technology, the number of digital image is growing rapidly. How to find the image data we need in the massive data base is attracted by more and more people. Now the Content-based image retrieval technique has been a important research subject.The Content-based image retrieval technique search the image in the image library by the content features. The result is similar to the target image. The traditional image retrieval algorithm gets the character of target image at first. By compare the character with the other image's corresponding characters, we can get the image similarity. Then according to the results, we can find the most similar image with the target image by select the image in the character base.In fact, there are the follow problems. The matching character results can't reflect the user's semantic requirements completely. The ability of serial search is poor in the image massive database. In a view of these problems, the thesis give the following improvements.(1) the thesis choose the more properly HSL color space .Then it divide the HSL space unequally which can reduce the dimensions and information loss effectively. At the same time it also carries the histogram's distance measure L1 distance with weight added.(2) The correlate mechanism of feedback is put and give an improvement which solve the contradiction between the increased feedback image and the users' eye strain.(3) The RPCL algorithm is put into the CBIR. With the improvement to the algorithms' defects, it improve the retrieval speed and the problem of how to determine automatically grouping number.In order to test the improving of the algorithm , the thesis build the corresponding experimental system. In the experiment of correlate feedback of image retrieval, the thesis's algorithm is compared with other algorithms which based on the accuracy , searching complexities. By the experimental data , the algorithm of this thesis has more superior retrieval performance.