Based On SVM To Classify And Retrieval The Image Information

The technique of content-based image retrieval (CBIR) is widely used in digital library, network information security, preventing crime, knowledge property right, medical treatment diagnose, geography information system, remote sensing and other areas. In recent years, The technique of CBIR has been a hotspot question of computer and interrelated subjects. It is also a foreland research problem.In this thesis, the methods about how to extract the color feature and the texture feature of image were researched. On this basement, SVM (Support Vector Machine) was used to retrieval the image. At last, the relevance feedback algorithm was adapted to enhance the retrieval efficiency.In order to extract the color feature of image, the image was changed from RGB space to HSV space. Then PCA was used to extract the color feature. This method effectively reduced the number of dimensions of the feature and the size of the feature database was decreased. According to the characters of wavelet, the second-generation wavelet was used to decompose the image to three levels, and then entropies of different levels were calculated. The entropies were regarded as the texture feature. This method made the number of dimensions of the feature was effectively reduced and calculation decreased.Based on the feature extraction, SVM was used to classify and retrieval the image. And the relative algorithms were given. The experiments proved that comparing with the algorithm before, the retrieval algorithm enhance the accuracy. At the same time, the effects of different kernel in classification were researched and the retrieval speed was researched, too.There was a great gap between the image feature extracted by computer and the meaning known by person. The retrieval result was not satisfied to people. In order to reduce the difference between them, the relevance feedback algorithm was adapted, which made the retrieval result meet the customers'need. At last a new good-sized CBIR system based on SVM was finished, in order to validate the validity of algorithm and key technology presented in this paper.