The Algorithm Research And System Implementation Of Content-Based Image Retrieval

The rapid development of multimedia technology has brought tremendous changes to the way people work and live. Because the dramatic increase in image data, the image retrieval technology came into being and become a research hotspot. Because of the huge working of manually annotation, the traditional method for information research based on keyword is hard to meet the requirements of users. Content-Based Image Retrieval (CBIR) is a kind of technique for retrieving similar images based on automatically extracting low-level features such as color, texture, and shape etc.Firstly, we mainly discussed the technical background of CBIR, some key technologies and the present and the future of CBIR, and analyzed systematically the low-level feature extraction of image. Secondly, we studied deeply the color feature of image and used the color histogram in different color space as feature to retrieve image, then selected the more effective method by comparing the experiment results. We choose the histogram in HSV color space as the color feature. Thirdly, we discussed the methods for extracting texture feature of image in more detail. Traditional methods include co-occurrence matrix, Fourier and the wavelet transform which has been widely used in image processing. There are a lot of researchers who extracted the texture feature by using wavelet transform. Although wavelet transform was an effective method for extracting feature, it has limitations such as not having translation invariance and limited direction selectivity. In this paper, we analyze the drawbacks of wavelet, and then extract the texture feature by using of Gabor, wavelet, rotated wavelet and DTCWT. After analyzing the experiment result, we choose the better method for extracting texture feature. At last, a retrieval method which combines color with texture feature is proposed in this paper. The algorithm considers the image color information, but also takes full account of the texture information. The results show that the effect of the combined feature is better than the single feature.At last, we use the Visual Studio2005as our IDE and C++language to implement the image retrieval system which based on MFC Document and View framework. In this system, the method for extracting texture feature is what we proposed in this paper, and the color feature is represented by histogram in HSV color space. We also choose the Corel image database as the experiment database. The experiment results show that the method proposed in this paper can improve the efficiency of image retrieval.