Theory And Method Research On Content Based Color Image Retrieval

With the rapid development of multimedia technologies and the growing popularity of Internet technology, the number of images coming from all areas of society is fast growing, image storage and management is an arduous task. How to fast and accurate find the images from the database become an urgent problem, so content-based image retrieval (Content-Based Image Retrieval, CBIR) emerged as an important research field of multimedia information Direction. This paper focus on content-based image retrieval core issues related to conduct in-depth research, the main contents include:(1) In order to avoid the problem that single visual feature can not fully describe the image content, it presented a based on color edge integrated image retrieval algorithm. First, using the detection operator Canny to extract the original color image edges; then construct three histograms which fully reflect the image edge contour information, that is: the color edge-weighted color histogram, angle histogram and gradient orientation histogram; finally, calculate the similarity between any two images and return search results. The algorithm can improve the retrieval efficiency.(2) In order to solve the semantic gap between image visual features and high-level semantics, it presents a based on the EM parameter estimation integrated support vector machine algorithm for image retrieval with relevance feedback. The algorithm first constructs AB-SVM classifier to solve the instability and the optimal SVM hyperplane bias problem; then construct RS-SVM classifier to solve the overflow problem; Finally, using EM parameter estimation method to make AB-SVM and RS-SVM integrated into a more powerful classifier for image classification. The algorithm significantly improved the traditional content-based image retrieval efficiency.(3) It presents a based on features reconstruction relevant feedback image retrieval algorithm. Firstly, the image feature is mapped to a high dimensional kernel space; then using the positive samples orthogonal complement component reconstruct sample images and test image features; finally, construct the image classifier with the new image features. Considering various defects of training samples, the algorithm reconstructs the image features and effectively improves the classifier performance.