Region Semantic And Vision Feature Based Relevance Feedback Image Retrieval

With the development of the technology of network, communication, multimedia and database, the multimedia information increase rapidly, the digital images are used more and more extensively. Nowadays, under the rapid expansion of digital images, how to organize, express, manage and retrieval the giant images have been a very active research field all the time. Under this circumstances, CBIR come into being and develop.However, because images possess abundant semantic information and complicated vision features, it is too hard to connect semantic information with corresponding vision feature of the multimedia target in present computer vision technology, which leads to the CBIR to be difficult to fulfill the request for practical application at accuracy of retrieval. The problem is the semantic gap between vision features and semantic information. This paper do some research on region and vision feature based image retrieval with relative feedback .The paper analyzed and summarized the fundamental, key techniques and performance evaluation of CBIR. In order to take region feature, we use an algorithm for color image segmentation, based on color and spatial information. Besides, a new region distance is proposed by this paper.In the prototype system, 72 bin color histogram and moment were chosen as region feature, and we use the IRM measure to calculate the distance between two images. A method is proposed to adjust region importance in IRM.At last , key word networks are established with vision feature. Relevant feedback is also applied in the system, it enable to catch the users'query intention by adjusting its similarity criterion automatically; Also, it pass the annotations for the relevant images, update weights between key words and images and fill the semantic networks. The best result will be presented for vision feature and semantic network can cooperate properly. A new method is proposed to update weights.