Studies On The Techniques Of Content-based Image Retrieval

With the coming of the era of information, it has increasingly become a problem to search knowledge in the information explosion. There is not only large amounts of normal data and text information, but also filled with more and more image informa-tion. These images are more difficult to handle and retrieve than text and normal data. Although there are some commercial search system already and they have been able to respond to the user’s text retrieval requests well, but the natural "semantic gap" exists between the image data and text information, and it makes text-based image retrieval system can not get satisfactory results. With these problems, generate content-based image retrieval system and related research areas.Content-based image retrieval system usually contains feature extraction, simil-arity matching, multi-dimensional indexing, user search interface and so on. In this paper, my studies focus on the low-level image feature extraction algorithm. Color features, shape features, the barycenter of color region and spatial distribution of image color information are improved and integrated. The color features is provided by the color histogram in RGB space, it is the basic of other features. Shape features is an estimated value by the largest bounding rectangle of color region.it can keep the rotation invariant of spatial feature. In this paper, spatial distribution features of the image is described by the Barycenter-vector. This method can be not only to describe the spatial distribution of color regions in the image, but also have the rotation invariant and scale invariant properties, the spatial distribution information extraction is also more flexible. The experiments show that by combining the above methods can achieve very good retrieval results in low-dimensional features.In this paper, The experiment development of language is Java which is a cross-platform language. Mysql database is the feature database. Besides, there is a GUI user interface by Swing for the CBIR System. The system includes the user request-response model, image feature extraction model, similarity matching model and results product module. In addition, it also has some other functions, such as statistical accuracy and the rate of recall. Through this platform to verify the effectiveness of the BVH image retrieval algorithms.