Design And Implementation Of A Content-Based Image Retrieval System

With the rapid progress of multimedia and internet technology, the amount of digital images is increasing with an astonishing speed. In order to use the image information more effectively, we need some kinds of technology which can help us search images more rapidly and accurately. This is named image retrieval technology. While traditional Text-Based Image Retrieval can not meet the requirement of the application in this area, Content-Based Image Retrieval come into being.This paper focuses on general image retrieval. We constructed a region-based image retrieval system with relevance feedback. This system mainly includes four aspects: image segmentation, feature extraction, image matching, and relevance feedback. In order to describe an image in region set, we adopted a color image segmentation algorithm which uses color and spatial information to segment the images, and reduces the segmentation time used by adjusting the calculation time of color scatter. In feature extraction phase, we extract 15 features including color, texture and shape for each region. In the image matching phase, we combine the color feature, texture feature and shape feature together to calculate the region distance, and then calculate the region similarity. Finally, an integrated region matching algorithm is used to calculate the image similarity, and return the retrieval results sorted according to the image similarity. The system also allows users to evaluate the retrieval results, and re-determined each region's importance according to user's feedback. Then the users can have a new query.This image retrieval system is robust to the imprecise image segmentation. And along with the introduction of relevance feedback, it can reflect user's intention in some degree and improve the retrieval precision. Experiments indicate the effectiveness of our retrieval system.