Study Of CBIR In DCT Compressed-domain

With the development of information techonology, CBIR (Content Based Image Retrieval) is much concerned. Kinds of image compression stardards are presented, which save the load of storage and transmission, but make the process of images more difficult. One way of solving such problem is that process image data in compressed domain directly.In this paper, some research is done on the image retrieval in DCT compressed domain.First, a new method of fast retrieving JPEG images is proposed. The feature extraction is done directly in DCT compressed domain, using the characterization of DCT. Furthermore, instead of fully decoding, only partial entropy decoding need to be done. Therefore, the work of retrieving images is accelerated. Also, the efficiency of retrieving is satisfied, with certain robustness.Next, a new algorithm for retrieval of texture images in DCT compressed domain is proposed. The statistical feature of the texture images is computed directly from DCT cofficients, and is used for image retrieval. It is supported by theoretical analysis and experimental results that the approach acheievs good performances in terms of retrieval efficiency and effectiveness.And it is invariant to rotation.Last, an algorithm of image retrieval based shape feature directly in DCT compressed domain is proposed. First, according to the edge feature of DCT compressed image, the rough object shape is extracted in DCT compressed-domain to form a binary edge map. Then, to improve the retrieval performance, the similiarity messurement is redefined using the moment invariant. It is also supported by theoretical analysis and experimental results, this approach not only is fast, but also have a good performance.And it is rubust to rotation and translation.