The Research On The Key Technologies Of Image Database Retrieval

With the rapid development of the Internet and digital equipments, the amount of multimedia information being produced, stored and spread is increasing rapidly. Therefore, research on multimedia information management is attracting more and more attention. In this research area, Image Database system plays an important role because of its foundational status of video applications as well as its direct employment in many important applications such as: digital library, digital museum, medical, geographical information database and military defense.Research on image databases can be divided into 2 stages. Early techniques were primarily based on the textual annotation of images. Obviously, annotating images manually is a cumbersome and expensive task for large image databases, and is often subjective, context-sensitive and incomplete. In order to overcome the difficulties of the early methods, Content-Based Image Retrieval (CBIR) emerged at 90's. Since the visual features of images can be extracted automatically by using the techniques of computer vision and image processing, it is much practical for large scale image database to be implemented based on CBIR.CBIR is a core technique of image database system. The main obstacle facing content-based image database is that the retrieval effectiveness is not satisfiable. Therefore, the difficulty in improving retrieval effect has remained as the main problem preventing large-scale image database system from practical applications.The main topic of this paper is about using long-term learning strategy within RF to improve the retrieval effect of image database. A retrieval framework based on RF log analysis is proposed. Based on this framework, some new retrieval methods are presented from the aspects of user model, RF pattern and the efficiency of the use of RF records respectively.The main work of this paper is as follows:· A retrieval framework based on RF logs analysis This paper proposes an image retrieval framework employing long-term learning strategy. In this framework, the Collaborative Filtering method is adopted to perform the analysis of RF records. An experiment prototype system is implemented to evaluate the performance of this framework. Experiment result shows that the effectiveness of the image retrieval can be improved apparently compared with conventional relevance feedback method.· A retrieval method based on the user RF sequence pattern analysis Along-term learning method based user model of CBIR and the concept of RF sequence pattern are proposed. According to the new user model, an image retrieval method based on the user RF sequence analysis is proposed. In this retrieval approach, Edit Distance is employed for the similarity measure between RF sequences in the process of RF sequence analysis.· A retrieval method based on the fuzzy clustering of RF recordsand information filtering The proposed method is based on the semi-supervision fuzzy clustering of the feedback records with information filteringBoth the user's relevance evaluation and the corresponding query images of the records are used to predict the semantic correlation of the databases images and current retrieval. The merits of our method are as follows: 1. more semantic correlation information can be obtained using the information filtering; 2. with the clustering of feedback records, the efficiency of the analysis of feedback records can be improved.· A retrieval method based on multiple sub-clustering centroids with relevance feedback The performance of relevance feedback technique in image retrieval is not stable in many settings, and more feedback information does not ensure more improving in effectiveness. To address this problem, we use unsupervised fuzzy clustering method to classify the user's relevance feedback images into several sub-classes, and each sub-class center point will be used to modify the original query so as to compute the score of each image by the fuzzy set rules.