Research And System Implementation On Content-based Image Retrieval

Content-Based Image Retrieval is a technology which retrieves image on the basis of the content feature of images. Its main idea is to analyze the content feature of images in order to find the images with similar content in image database. As a comprehensive technology which combines several technologies, such as information retrieve, image processing, computer vision, artificial intelligence, CBIR has become an urgent research area broadly in multimedia age.This dissertation makes extensive and deep research on CBIR technology. The main content of the dissertation is summarized as follows:1. This dissertation deep researches the general algorithm like underlayer vision feature extraction, similarity measurement among images and criterion of algorithm performance, and introduces MPEG-7 which is an international information description criterion applied in CBIR.2. An improved approach for image retrieval based on MPEG-7 texture edge and generalized co-occurrence matrix is proposed. The new approach constructs five co-occurrence matrixes with five directions. And it not only considers the spatial correlation information but also takes the direction information of images. As experiments show, the method has high precision ratio and recall ratio.3. A new retrieved algorithm which combines improved MPEG-7 dominantcolor with edge histogram is proposed. Firstly, the algorithm iteratively extracts initial cluster number and cluster centre. Then it extracts dominant color of images using GLA clustering algorithm to combine with edge histogram. The new algorithm would improve retrieve performance.4. A CBIR system for testing retrieval algorithms is developed to test the algorithm proposed by this dissertation.