The Research Of Image Retrieval Based On Color And Texture Features

With the development of technology, the images on the internet are growing explosively. As the number of images is becoming larger and larger, different kinds of images mix together. It is difficult for the users to find the information they need. In order to realize the efficient management of image database and improve the efficiency and accuracy of the retrieval, the content-based image retrieval (CBIR) is proposed and becoming a focus of discussion at home and abroad.The key techniques of CBIR mainly include feature extraction, similarity measurement, indexing techniques, and relevance feedback, evaluation criteria. In this paper, we focus on the research of feature extraction. The main contents of this paper are summarized as follows.Firstly, the traditional color histogram can only represent the statistic distribution information of the pixel color in the image. It loses the spatial information. In order to solve this question, we study a method of image retrieval based on block-histogram. The image is divided into several different sub-blocks and the color histogram of each sub-block is extracted. The similarity between the color features of corresponding sub-blocks in two different images is calculated by the Euclidean distance. Then, the real similarity between two images can be obtained by weighting the above distance. On the basis of traditional color histogram, we introduce the spatial information in our method. Compared with the traditional color histogram, the experimental results show that our method can achieve higher precision than the previous method.Second, the original Local Binary Pattern (LBP) descriptor has some demerits. It is sensitive to the noise. In addition, sometimes the different structural patterns may have the same binary code. These two disadvantages reduce the discriminability of LBP descriptor inevitably. In order to solve this question, in this paper, we propose a method of image retrieval based on block local binary pattern. The original LBP is improved by calculating the LBP using the average local gray level of a3×3local area instead of the gray value of its central pixel. The image is divided into several different sub-blocks, and the LBP feature of each sub-block is extracted using the improved LBP descriptor. Similarity between the LBP features of corresponding sub-blocks in the two images is calculated with the measure of Chi-square. Finally, the similarity between the two images can be obtained by weighting the similarity calculated above. Compared with the original LBP descriptor, the result of experiment shows that the method proposed in this paper has good precision.Third, we study a method of image retrieval based on multi-feature integration. Single feature couldn’t express the complex content of image comprehensively. Retrieving with the single feature of image, the results couldn’t always satisfy the needs of the users. On the basis of the two methods mentioned above, we proposed a method of image retrieval based on the combination of blocked color histogram and local binary pattern. The experimental results show that the image retrieval method combining color feature and texture feature together has better retrieval precision.