Research On Image Retrieval Combining Semantic Feature

Content-based Image Retrieval still encounters much difficulty which is generated by the semantic gap between image semantic features and the lower features, resulting in the fact that the extracted content features still mainly centered upon the lower features such as color, texture and shape. Therefore, in a long run, it is still an unsolved problem about how to integrate with semantic features to achieve better connection between the lower physical features and the image content for effective retrieval.In this paper, a lot research work has been done around three points: how to abstract and index low_level feature of images, how to retrieval images in semantic level, and how to fill the semantic gap by relevant feedback technique. Based on the research work, an efficient and practical image retrieval system is built which integrated the advantages of these techniques. Refer to process of understanding and analyzing of human, image being divided into background and object, draw corresponding visual feature respectively as parameter of retrieval, carry out image retrieval. Because of different subjective perception of users, it can accept user's participation as appraise for retrieval result, resort to it adjusting the weight of each feature parameter, make the retrieval step by step going on according to the requirement of user, up to accord with the needs of user on vision and semantic. Concept and characteristic of CBIR technology have been elaborated, extract algorithm for each feature and image matching were discussed and their advantages or disadvantages are analyzed from the application view. First of all, after researching shape feature-derived and feature matching, the paper bring forwards a new improved feature matching algorithm on the basis of Su Yang descriptor, made image retrieval holding a very fast matching speed and a high matching precision. Secondly, key word networks and low-level feature table are both established for images, which realized double indexing; relevant feedback is also applied in the system, and it pass the annotations for relevant images, update weights between key words and images and fill the semantic networks,thus the system can study in long-term. Users can provide their query requirement in several different manners, the best retrieval result will be presented for the low-level feature and semantic network can cooperate properly. The experiment shows that retrieval accuracy of the system is above 70% in few steps of relevant feedback (five times in average), and the performance could increase steadily with it been used.