Research On Image Retrieval Method Based On Texture Feature

With the development of information society, the emergence and extensive use of the mass storage facility and digital equipment, together with the popularization of multimedia and network technology lead to the vast image data increments. The content-based image retrieval (CBIR) emerges and has been focused upon in order to quickly and exactly query images that they need from the image database. CBIR that is different from the traditional text-based image retrieval is a fuzzy inquiry technology in fact. Through extracting automatically and intelligently visual features from each image as its index, we find the feature space and similar query images so as to realize the automation, intelligence and humanness of image query and management. In the CBIR system, extraction feature and similarity matching become very important. The existing color, texture and shape feature extraction, texture feature is widely applied in the Image Retrieval because it can do better showing the characters of the image. Therefore, the paper mainly study the texture-based image retrieval.This dissertation mainly research and analyses the key technology of content-based image retrieval, and also compare it with typical retrieval algorithms. In addition, an image retrieval prototype system has been developed.The main contributions of this dissertation are summarized as follows:Firstly, The paper makes some research on various texture retrieval algorithms and the texture feature description on the basis of which a new algorithm of adaptive weighted LBP has been designed. The question we consider is as follows:1) It is known that the window size in traditional LBP algorithm is fixed. Our proposed algorithm which is able to adaptively analyses window size due to combine the Tamura texture coarseness with traditional LBP algorithm realize better texture analysis performance..2) There are different meanings due to different pixels location in different area of image. We get a good retrieval effect through calculating weights of each pixels by Laplace operator, that is our proposed adaptive weighted LBP algorithm.Secondly, it came to the conclusion that the key to influence the performance of image retrieval is the feature extraction of image and the similarity between two images by analysis and discussion on the theory of content-based image retrieval system, general frame, key technology, performance evaluation and relevant feedback of content-based image retrieval system.Finally, A texture-based image retrieval prototype system has been designed and developed using Matlab7.0. The prototype system provides certification for various classical algorithms and the proposed algorithms in this paper.This paper reports the study on content-based image retrieval. On the basis of a variety of typical retrieval algorithms, we put forward my own ideals and propose a new algorithm of adaptive weighted LBP.