Research Of Medical Image Retrieval Technology Based On Multi-feature

With the rapid development and the progress of technology, the bulk of daily markettrading, stock futures in the financial area, the medical industry has accumulated a lot of data,a lot of people Analysis these statistics from accumulated data, find it difficult to minevaluable information longitudinally. Thus, data mining algorithms is proposed, data miningcan find useful information in complex data effectively.In the health industry, the hospital has accumulated a large number of images. In2010,a hospital image storage is2T on average, and every year by double-digit increments. In theaccumulation of medical image, in is difficult for medical staffs to find pathology similarmedical images. Traditional image retrieval could not meet the demand of the growing, weneed develop an effective medical image retrieval system, can make up for the traditionaltext based image retrieval problem in content based image retrieval. Therefore, cbmir have agreat potential.In order to solve the problem that a single feature could not express complex medicalimage features in the image retrieval effectively, improving the existing medical imageretrieval method, pairwise constraint method of semi-supervised learning features selectionwith an ant colony clustering method which combines shape features and texture features isproposed in the paper. Firstly, combine texture features of the gray level co-occurrencematrix and shape features of Hu moment invariant, setting up a multi-feature of medicalimage database, then choosing appropriate medical image feature using pairwise constraintdimensionality reduction. medical images with high similarity cluster together by ant colonyclustering algorithm. Finally, medical images in the dataset are retrieved by using similaritymeasuring algorithm of weighted Mahalanobis distance, which optimizes the recall andprecision ratio of retrieval, boosting the stability of classify and stability in the retrieval. Themethod, which combines the advantages of two features and optimizes the effective ofretrieval, overcomes the problem that a single feature could not express complex medicalimage features, attaining the expecting result. This system can realize the dynamic medical image retrieval based on original retrievalplatform. Refer from medical image retrieval technology of Guo Jinge and feedbackadvantages of Liu Wei, the image retrieval system based on multi features. Researchprogress are as follows:Extract texture feature and shape feature of medical images, build a medical imagedatabase. Test of the gray level co-occurrence matrix, Hu invariant moments, two kinds offusion algorithm of search results. A preliminary conclusion: single feature can not meet thedemand for medical image retrieval, two kinds of fusion algorithm, the initial retrieval resultsthan single retrieval effect is good, but need to set the weight of two kinds of algorithms.