The Research Of Target Region Based Image Retrieval System

With increasing of image resources, the management of them has become more important. As well, image retrieval has become the hot spot in research at home and abroad. Based on Text-Based Image Retrieval (TBIR), Content-Based Image Retrieval (CBIR) has been raised. It overcomes many disadvantages in TBIR, such as wasting large amounts of human resources and easy to influence by some subjective factors. In CBIR, retrieval based on global features is relative simple, fast to calculate. But because of ambiguity, retrieval effect will be affected by background and other objects in the image. Retrieval based on local features can reduce this disadvantage in global retrieval. But the retrieval technology recently still has a lot of defects, and can't accurately extract target object region interest by user.To locate the region of the object needed in image accurately, we complete the following works. At First, after dealing with images in database, we raise a multi-color object's region extracting algorithm using the neighborhood list based on adjacent property of regions. Second, aiming at color region, we extract the area ratio of main color region blocks as color feature and the texture features based on co-occurrence matrix of color distance. Third, synthesizing the color feature and texture features, we determine and improve a similarity calculation algorithm suitable for the object region extracting algorithm before, and do the inter-image similarity matching. When retrieving the images, the object region features has been used instead of the global features to calculate the inter-image similarity. The experimental results showed that this method can accurately locate the interested multi-color object in the image. During similarity matching, much influence, caused by the background and other objects, has been excluded in the similarity calculation. And retrieval accuracy can be improved, too.