The Research Of Content-based Image Retrieval Technologies

With the rapid development of Multimedia and Internet techniques, people can get more and more image information. The application and spread of image are extended, however, it is difficult to organize and manage the increasing image database. It is urgently to build a high effective image management system. These requirements drive the research and development of image retrieval technology. As a new technology, content-based image retrieval (CBIR) has been a hot topic.CBIR includes low level visual feature and high level semantic feature. Because of difficulties for abstracting semantic feature, most of researchers based visual feature to develop some systems. But the question of Semantic gap appeared because dissymmetry was exited between visual features and semantic features. Experts have indicated that the resolution of Semantic gap will depend on the breakthrough of relevant technologies.For narrowing Semantic gap, we commonly research three key technologies as bellow: Firstly, abstraction and description of visual features, which is the basic of CBIR; Secondly, a good algorithm can reduce complexity of time and space, and also have a high performance; The last is the technology of relevant feedback, which recur to the participation of users toimprove the result.Key technologies of content-based image retrieval have been deeply studied in this paper. The research background, the importance of the issue, the up-to-date applications are briefly introduced. Then some universal methods of feature abstraction are described. Aim at solving the low precision and semantic gap, a new relevant feedback method is presented. The new method only need user rank the last retrieval according to his interest only and can avoid labeling the key words for images. After calculating the R_norm between system outputranking and user own-wish ranking, system can adjust the weight of features automatically. The experimental results show that the new relevance feedback mechanism outperforms Rui feedback method. It can get more coincident result to user, and may reduce computing complexity in a certain extent.Our study may improve the performance of image retrieval system. But only relevanttechnologies as pattern recognition, image understanding, artificial intelligence etc. getbreakthrough, Semantic gap can be absolutely resolved.