Studying On The CBIR System Based On NSCT

With the rapid development of the internet and the multimedia technology,the application of the image is more and more extensive,the research of the Content-based Image Retrieval (CBIR) becomes most active one because of its practical significance.The thesis expatiates the existing feature extraction algorithm based on the color,texture,and shape features and analysises the disadvantage on the base of describing the image feature and its sorts briefly.The thesis reviews the history of the Multiscale Geometric Analysis(MGA),the theory,the properties and the implementation of Nonsubsampled Contourlet transform (NSCT) are chiefly introduced in this paper.The core of the paper is a algorithm that's feature extraction with NSCT .We use this algorithm to construct a CBIR system,which combined with the energy feature matching technology based on Euclidean Distance. Simulating in the Matlab7.0 with the Brodatz image library,which concludes 112 grey pictures(640×640),the results show that the system provides very good performance when the retrieved images aren't shape-changed.Then Log-polar transform is used to improve the system in order to make it rotation and scale invariant.Our results indicate that the improved system is robust and yields satisfying performance when the retrieved images is rotated or scale-changed or both of them.