Color-Based Image Retrieval Methods And System Implementation

With the rapid development of computer and technology of the network, especially the popularization of Internet since 1990s, the computer has gradually become the necessity to people's consumption and daily life. Meanwhile, the amount of information in the Internet increases at a dramatic speed, and Internet has become a genuine information storehouse. At the time users are enjoying great convenience offered by rich and colorful media information, they are also facing the problem that how to find the necessary information quickly from the abundant information storehouse. When refers to searching and seeking the multimedia data in the database, we mention that early method is to set up a text descriptive information, such as key word, as an index for every image in the database, through people's observation of the picture, then people carry out the search on the basis of the key word. But as the capacity of the image database is larger and larger at present, the difficulty of such traditional method based on text key word becomes very outstanding. In order to overcome the limitation of the traditional searching method, Content Based Image Retrieval (CBIR) technology was aroused at the beginning of the 1990s. As an effective way to solve the problem of the search for multimedia information, Content Based Image Retrieval technology has a wide applied prospect. This paper analyses and summarizes the main technology of CBIR, and also discusses the conception, methods, some acquired achievements and future research trends of CBIR. The systematic frame, the search procedure, the characteristics collect technology, the semantic characteristic, and the performance evaluation of CBIR system are introduced. The key to implement the technique of CBIR is extracting feature which present the content of image. As an important visional information of image, the calculation of color characteristic is simple and steady, which has been broadly used in CBIR . This paper have discussed three key questions on how to make use of the color characteristic: the expression of color, the abstraction of color characteristic and the measurement of likeness based on color. Through the discussion and comparison to several common color spaces, the writer carry out some research on color space choosing technology in CBIR. On this basis, the extraction technology of the color histogram characteristic is especially introduced. Finally this paper introduces the common color similar measured function in CBIR, and discusses in detail the image retrieval algorithms based on color histogram in CBIR. This paper considers the advantages and disadvantages of traditional histogram and Partition-based histogram, propose a new method based Partition-overall histogram. The basic thought of Partition-overall histogram is to divide the image space according to a certain strategy, and then calculate color histogram of each block as the color feature of this block. Users choose the blocks that contain important space information, confirming the right value. The system calculates the distance between the corresponding blocks that users choose. Other blocks merge into part overall histograms again, and the distance should be calculated. Then accumulate all the distance as the real distance between two pictures. The less distance there is, the less visional difference they have Partition-overall histogram neutralizes advantages of two methods above, by choosing blocks makes the feature contains more spatial information which can improve performance; The distances between part overall histogram make rotating and translation does not change. To improve the usability of the system, we introduced the technique of relevance-feedback. As far as we know, the features adopted by most CBIR system contain color, texture, shape and relation-based shape which gained more attention these years. All of them are low-level features and differ from the way brain describe objects. So, the recall rate and precision rate of current CBIR system are not very satisfactory. Relevance-feedback is the mostly used technique to refine the query result in CBIR system now. Since it was proposed by Recchio in 1971, it has been adopted by many information retrieval systems. It has been proved that the use of relevance-feedback method can improve system performance significantly. In our system, relevance-feedback is applied in two ways: re-weighting based and query...