The Research On Content-based Image Retrieval Algorithm

With the rapid development of network and multimedia technology, people will get more and more images. How to search the images that users need from the huge image database has become an important issue in the field of image application and needs to be solved. Text-based image retrieval is the traditional way to retrieve the similar images that users need in the image database. This kind of artificial method not only wastes the manpower of large quantity, but also depends on the people’s subjectivity. Content-based image retrieval can overcome the disadvantages of this method and has become the focus of image retrieval technology. Although the number of image retrieval methods is increasing, perfect the theory and pursuit better result of image retrieval are always the goal of the researchers.First of all, the research background and significance of image retrieval are introduced, and the key technologies of image retrieval are simply presented in this paper. Then, we mainly devote ourselves to studying the color based and texture based image retrieval methods. The main works of this paper are as follows:(1) In color based image retrieval, the color histogram is used for feature extraction. According to the type of color space, this paper studies the color histogram in RGB color space and HSV color space respectively. In RGB color space, this paper proposes a new method for image retrieval by using the histograms of images in R, G and B channels as color feature. In HSV color space, as the global color histogram lacks the spatial information of color, this paper studies an improved way by dividing the image into several blocks and allocate different weights for them. We compare the different methods by experiments and the experiment results show that the improved way performs better than the previous method.(2) In texture based image retrieval, this paper studies two methods to extract texture feature, one is wavelet transform, another is co-occurrence matrix. The co-occurrence matrix includes gray level co-occurrence matrix and its improvement method called texton co-occurrence matrix. This paper analyzes these three methods by experiments and compares the retrieval results of gray co-occurrence matrix with texton co-occurrence matrix. The experiment results show that the retrieval result of texton co-occurrence matrix is better than gray co-occurrence matrix.(3) In color and texture based image retrieval, this paper studies two ways to integrate color and texture features for image retrieval. One way is to construct color and texture feature vectors respectively first. Then, calculate the similarities of color and texture features respectively. Finally, use the weighted addition to the similarities of color and texture features for image retrieval. Another way is to construct a feature vector that has both color and texture features. In the first method, this paper proposes a new algorithm for image retrieval by combining the weighted block color histogram in HSV color space with texton co-occurrence matrix. In the second approach, this paper improves the retrieval method called MTH by adding the process of image segmentation. The image is divided into nine equal blocks and the feature of each block is extracted to construct feature vector for image retrieval. At the end of this part, this paper compares our method with other existing methods by experiments, the experiment results show that the methods we proposed in this paper are more effective and accurate.