Image Retrieval Algorithm Based On Fusion Of Ct Feature And Texture Feature

Content-based image retrieval has become a hot retrieval technology, but in the feature extraction and similarity measure, there are still some technical issues to be resolved. By studying the existing image retrieval algorithm, this paper we propose an image retrieval algorithm based on fusion of Contourlet feature and texture feature. Firstly, on the basis of images in the database to make the size of block transform processing on the extracted image and the standard of the average energy of the difference between these two converted statistical feature to generate a feature vector. And then do “uniform” Local Binary Pattern for the image is preprocessed, and then Gray Level Co-occurrence Matrix treatment, to extract angular second moment, contrast, correlation and entropy of the four feature vectors to generate statistical texture feature vectors. Finally, the integration of these two features as a feature vector images. Canberra distance is adopted as similarity measure. Complete the image retrieval experiments.This paper are focuses on the algorithm in Contourlet decomposition level, LBP operator to select the sub and GLCM algorithm parameters, and select the similarity measure function were studied experimentally determine the optimum parameters of the algorithm. In addition, also in the different images in the database application algorithm for image retrieval experiment achieved higher average retrieval rate, verification algorithm good generalization ability. Also on the MIT image database retrieval algorithm comparison with other experiments, the algorithm search results reached 89.68%, higher than the rest of the algorithm, to verify the effectiveness of the algorithm. Specific work schedule is as follows:(1) Select the parameters of the algorithm to determine the best algorithm. We research the Contourlet decomposition level, the radius and the number of pixels LBP operator, GLCM algorithm distance and similarity measure of image retrieval methods on average precision rate in the typical image database. Through analysis discovery that when we choose(0,2,2,3) Contourlet transform level, with a radius is 1, pixel is 8 of the "uniformed " LBP operator, distance is 1 of the GLCM, Canberra distance similarity measure to carry out image retrieval the retrieval result is the best.(2) Research the generalization algorithm. When used in the MIT image database, Brodatz texture image database and image library Outex50 proposed algorithm for image retrieval experiments, the average retrieval rate algorithm are up 89.68%, 80.03%, 81.61%. Experimental results show that the algorithm has good generalization ability.(3) Research on the effectiveness of the algorithm. The proposed algorithm is compared to the algorithm of image retrieval that integration of LBP and GLCM texture feature extraction, integrating texture statistical features and Hu invariant moments in Contourlet transform, and based on the LBP operator image feature extraction algorithms, four experimental study comparing them to the average rate of the image retrieval respectively 89.68%, 87.38%, 83.39% and 82.86%. The analysis found that the proposed algorithm can get higher the average retrieval rate, show the effectiveness of the proposed image retrieval algorithm.