Study On The Algorithm Of Color Image Retrieval Based On Fuzzy Theory

In the multimedia technology age, network is full of a large number of image information. How to query image with rich content quickly and efficiently has become a hot research field in academe, and content-based image retrieval (CBIR) technology emerges as the times require. Due to fuzziness of human being's perception, image retrieval facing user should accord with mode of thinking of human brain. Fuzzy mathematics is exactly the powerful tool to simulate human brain thinking, so fuzzy theory for analyzing, researching and realizing image retrieval is proposed in this paper.Firstly, this dissertation analyses the key technology of image retrieval such as image feature exaction and similarity measure between two images, then the basic principles of fuzzy theory are introduced. After that, image retrieval algorithms based on fuzzy theory are studied and brought into effect, and experimental results show the relatively favorable precision and efficiency of the proposed method. The main construction of this paper is summarized as follows:1. A new color histogram and a method of similarity measurement are proposed in order to overcome some disadvantages of traditional histogram. Firstly, MPEG-7 vision content-based color is adopted as the feature in similar color area, and the frequency of each color after asymmetrical quantization of pixel colors is calculated. Then extract main color feature as eigenvector in image database. Finally, based on the fuzzy set theory, a distance function can measure similarity of color histogram.2. Fuzzy Hamming Distance (FHD) is presented, which is extension of Hamming Distance applied in fuzzy theory, considering both the difference between features (the feature is color in this paper) and degree of difference. After image segmentation, we extract HSV color features from each image block and compute similarity measurement between query image and database images with FHD. This algorithm is applied to a professional database named Corel, bridging the great gap between high-level concepts and low-level features preferably. Compared with FUZZYCLUB system from Ruofei Zhang, this method demonstrates a promising retrieval performance. 3. During the image color quantization process, color locating in quantization boundary is fuzzy. Hereby, we adopt membership function in HSI color space to describe color. Furthermore, position distributing and disperse degree in fuzzy subset of color features are also added to show space information. Because texture is the important information to describe image, classical gray level co-occurrence matrix feature is extracted. At last, we do image retrieval by combining color histogram, with color-spatial information and gray information.4. For the reason that color histogram only reflects the statistic character of color, retrieval scheme combining four image feature descriptors is presented. From the four aspects of color, texture, edge and space, histograms are extracted, then follows Fuzzy Hamming Distance (FHD) which is used as a fuzzy similarity measure. We have performed an experiment on a 1000 images professional database named Corel and our results show higher retrieval accuracy.