Based On Color Features Combined With Relevance Feedback Image Retrieval Technology

In recent years, an increasingly large number of multimedia database, the number of images in the database is growing all the time, the traditional Image Retrieval methods have been difficult to meet the needs of users, which makes the researchers put more attention to the Content Based Image Retrieval (CBIR), under the unremitting efforts of the researchers, CBIR has developed rapidly, has obtained the very good Retrieval effect, when people is retrieved, is no longer Based on text Retrieval, but using the underlying characteristics of images, such as color, texture and shape, etc. Among them, the most intuitive features belong to the color of the bottom of society, when people see an image, first perceived is its color information, so, in the study of content based image retrieval, the people is the most commonly used to retrieval based on color feature, and this kind of method can be used for any type of color images.In the extraction of color features, widely used global color histogram, compared with other approaches for image features, it has a big advantage is that it has a good rotation, translation and scaling invariance, use the more flexible when retrieving images, applicable scope is wide, and the calculation is simple., however, it is not very good expression of color space position information, that is to say, the color distribution is not the same completely, two images are likely to be their color histogram is the same, so visually different images could be retrieved completely come out, this is not the kind of people of hope. So, researchers put forward the thought of local histogram, this method first turn on the image according to certain rules to block, divided into several small areas, each area is extracted and then color histogram, and use appropriate similarity measure criteria, calculate the corresponding regions, the similarity between the weighted sum of the degree of similarity to the overall image. Chunked retrieval way through each block, the location of the relationship between contains spatial information of image to a certain extent, makes the retrieval results of global histogram had greatly improved than before. But, the simple image block and there is no guarantee that the image rotation and translation invariance.Therefore, this paper proposes a combined image is the overall method of color features, color features and blocking the first extraction image color histogram of whole, again the image region partition, by block weighted, combined with color features of image retrieval. This method combines image the whole and the block color distribution, which express the spatial information of image, and guarantee the rotating invariance and translation invariance. Global color histogram to make up for a partitioned by the local histogram were lack of rotation, translation and scale invariance, and block the local histogram is good enough to make up for the global spatial information distribution histogram have no, both are combined, has achieved very good retrieval effect.In this article, the degree of similarity between two images for the overall weighted partial similarity and similarity and blocking, and finally, in the retrieval and joined the relevant feedback technique, after finished the study of algorithm, this paper design and implement a retrieval system based on local image library. At the end of this article, the article has carried on the summary, and put forward the outlook for follow-up work.