Texture Image Retrieval Based On Gabor Wavelet Transform

The purpose of the research of content-based image retrieval is to realize retrieving images automatically and intelligently. The objects of the research are methods and technology that can help the user retrieve particular images from image database conveniently, quickly and accurately. Content-based image retrieval has been a pop research question both domestic and international in recent years. CBIR means retrieving image according to various features of the image contents directly. The research on CBIR focuses on extraction of lower-features and similarity-matching currently.As a basic vision feature of vision attributes of humankind, texture is widely distributed. Texture-based image retrieval is one of the most important methods of image retrieval. This thesis analyzed and compared several classical methods of texture feature extraction and some common similarity measure methods and some relevance feedback methods. The main research work is to lucubrate methods of texture feature extraction and relevance feedback. The main work of this thesis includes several parts as follows:(1)This thesis studied the key technology of CBIR, analyzed and compared the primary methods of image retrieval.(2)This thesis analyzed the characteristics of texture and the status of texture-based image retrieval and summarized the advantages and disadvantages of each method by comparing the existing methods of texture feature extraction and similarity measure.(3)Gabor wavelet was used to extract texture feature of images and three weights of Tamura texture features such as coarseness, contrast and directionality were added to construct the texture feature vector of image and then this method was analyzed in detail.(4)This thesis improved a conventional relevance feedback algorithm by importing an updatable feature database into the improved algorithm. With this algorithm the system can gradually embed the user's feedback information into the updatable feature database.(5)An experimental system was developed and experiments on Brodatz and Uni-bonn texture image databases were finished with our method of retrieval. The performance of the experiments showed that the method is very efficient.