Research On The Algorithms For CBIR Feature Extraction

Content-based image retrieval is a very problem that is worthy of studying in the field of information retrieval. Researching practical content-based image retrieval system and finding out the relationships between images have the important academic value and practical significance. Image feature extraction as a key technology of content-based image retrieval, the texture feature of the image is researched and ensemble retrieval is explored for many algorithms, and then an ensemble retrieval algorithm is proposed. On the basis of the analysis of the retrieval algorithms, a similarity measurement method is researched that suited to content-based image retrieval. Finally, the corresponding image retrieval prototype system is designed. The main contributions are included as follows:An ensemble retrieval algorithm that retrieving with the merits of various algorithms is proposed based on analyzing the existing algorithms based on image texture feature extraction. The main idea is:firstly, the energy of the image is extracted as its feature vector; secondly, the weight of the middle area of the image is increased by partition-weighting due to the main information of the image concentrated in the middle area; finally, the ensemble retrieval algorithm for the texture image based on wavelet transform is proposed by using the weighted average of the algorithms based on the retrieval credibilities of the image retrieval algorithms based on wavelet transform. Precision and sorting evaluation method are adopted as the evaluation criteria for retrieval algorithms. Experimental results show that the algorithm proposed has better retrieval results.A similarity credible measurement distance is proposed after finding that the selection of similarity measurement method will affect retrieval algorithm performance during researching image feature extraction algorithm. The distance put forward mainly considers that different measurement methods for retrieval algorithm have different effects that the retrieval effectiveness. At the same time, some properties of the similarity credible measurement distance are analyzed in depth and the proofs are given. Experimental results show that the retrieval effectiveness of the retrieval algorithm is improved by the measurement distance proposed.A content-based image retrieval prototype system is designed and realized that based on the above researches in Matlab7.0. And this system realizes the content-based image ensemble retrieval that can contribute to the subsequent in-depth research.