Image Feature Description In DCT And Application In Image Retrieval

With the fast development of PC technology, Multimedia and network technology, the digital image becomes main information carrier gradually, and gets broad application in all professions and trades, and various types of databases are formed during the course. For the database managing and using an image more well, research on the feature extraction in DCT domain and application in image retrieval is representative in image retrieval technology.Available studying of image retrieval in DCT domain is summarized, according to fundamental optic characteristic, feature description method in DCT domain are classified by: feature description of mean gray value in DC coefficients, statistic directional and multi-resolutional characteristic in AC coefficients, statistic texture description in DCT compressed-domain. theory, property and application about SVD have been introduced.Algebra feature extraction-SV vector in DCT domain is proposed in image retrieval. This arithmetic combine statistic directional and multi-resolutional characteristic with SVD. First, rough texture matrix I1 ,I 2 is obtained, then SVD is applied respectively in the matrix I1 ,I 2, and partial or all of SV vector according to the characteristic of images is used to image retrieval. For improving the algorithm against intensity change, it carried out improved Euclidean distance.In the end, experiment is carried on worked-image database, texture-image database, similarity -image database. Experiment results show that SV vector is not against rough textured matrix's expression of original image and the proposed technique possesses not only the robustness against illumination changes, position rotation, sharpen, Gaussian, and noise of image, but also it has better performance on similarity-images and texture-images.