The Research On The Key Technology Of Content-Based Image Retrieval

With the rapid development of the Multimedia technology and Internet, the image data are sharply increasing. The image data are usually obtained from satellite system, monitoring system, science test, biomedical field and so on. For example, the earth observation system of NASA produces about 1T image data every day. How to get the needed image information from massive image data is an urgent and important problem. The traditional information retrieval methods which based on the value/charater can not objectively reflect the diversity of image content, and the data model, system architecture, query method and user interface based on the traditional methods do not have the capabilities of managing and retrieving image data. So, the studies on the method and technology of content-based image retrieval (CBIR) have important theoretical and practical significance.This paper investigates the content-based image retrieval method from five respects including system architecture of CBIR, image data model, orgnigation of image data, feature extraction and similar measurement. Firstly, the system architecture of CBIR which has wide application nowdays is used. Then a data model with image character based on the extended relation data model is proposed. In addition, the organigation method of image data is given, and the R-tree is been used as the index of image database. On the basic of the index method, a K-NN retrieval algorithm is presented by calculating the MinMax distance. In feature extraction, the color histogram is used to extract image color feature in YIQ color space; the 2D Gabor wavelet transform and texture co-occurrence matrix are used to extract image texture feature; K-Means cluster method is used to segment the image into regions, and get the image region feature. At last, a novel similar measurement method based on the most interested region is presented.In order to test the performance of the key technology presented by this paper, a prototype of CBIR system has been designed and implemented. The experimental results show that it has better recall rate and precision rate than the traditional CBIR systems.