Image Retrieval Based On Color, Texture Features And Relevance Feedback

Along with the development of computer and communication technology, digital image processing is widely used in many fields. Traditional information management and retrieval methods are no more appropriated for large image databases. In order to manage and retrieve large amount of images, the CBIR (Content-Based Image Retrieval) has emerged to be one of the hot research areas in image domain.In this paper, the key technology of CBIR is studied. And the main aspects dealt with this are structured as follows:Firstly, the color space (L~*a~*b~*, HSV, RGB) and the extraction of color features are discussed in this paper. The traditional histogram image retrieval has not encompassed the color space information .In this paper, a method which segment an image into six rectangular regions and the local accumulative histogram is discussed that is based on the L~*a~*b~* color space is proposed. The retrieval effects were contrasted through the experiment.Secondly, two texture extraction methods: cooccurrence matrix and Gabor wavelets, are compared. Experiments show that the frequency method as Gabor wavelets is better than the method which extract texture feature in space.Thirdly, this paper proposed the synthesis retrieval method on both color feature and texture feature of color image, the experiment indicated may obtain the good effect.Fourthly, as the weak correlation low level features and high level conception of images and different subjective perception of the system users, relevance feedback has been introduced to CBIR and the system learns the requirements of different users from the feedback procedures. This paper studies using the K-means clustering algorithm to classified the obtained image, submits through two phases to retrieve the image and adjusts the weight using the relevant feedback method. The experiment indicated that the retrieval could enhance the rate of accuracy immensely.At last, a prototype image retrieval system is constructed, which realizes the content-based image retrieval and feedback image retrieval.