Research On Content-Based Image Retrieval

Content-Based Image Retrieval system and several typical systems are firstly introduced in this paper. Then three characteristics of image, which are color, shape and texture, are studied respectively. During the study of color matching, HSV color model is chosen and some dividing methods for similar colors in HSV space, which are to quantify color in the color-based image retrieval, are proposed after studying different color models and different color features. The experimental results show that HSV color model has good effect on retrieval, and histogram can validate the total color information with regard to an image, instead of being sensitive to position change. Using color block retrieval approach, two methods are proposed to divide an image into some sub-blocks, which provide position information to some extent. Therefore, augmenting the weight of the ones concerning users is able to make it possible to enhance the intelligence and correctness of image retrieval. During the study of the shape-based image retrieval, statistical algorithm is firstly suggested, which is simple as much as easy to realize. In addition, a novel moment called polar-radius-invariant-moment and object shape description-Fourier descriptor based on two-dimensional polar Fourier transform is proposed for the object recognition and classification. The experimental results show that these methods have many good properties such as invariance of moving, rotating and scaling in space. During the study of texture matching, based on the algorithm realizing the image retrieval by texture characteristic of gray degree symbiosis matrix, the method to match texture is further proposed by incorporating gray structure patterns and texture statistical approach to construct grey-texture primitive co-symbiosis matrix as the characteristic of image. The experiment results indicate that the method can reduce retrieval time and improve retrieval precision. Finally, to attain much richer semantic from the image to apply the image retrieval, the retrieval algorithm by synthesizing color, shape and the characteristic of texture is introduced. The experimental results show that image retrieval method on synthesis features is much better than on single feature.In this paper, the retrieval ways mentioned above is programmed and realized by Visual Basic 6.0. And then the validity of the algorithms is analyzed. Although the experimental results show that the algorithms are efficient, there are still many problems that may be ignored in the paper and need improving in our further research and analysis.