Texture Image Retrieval Based On Spectral Graph

With the development of digital multimedia technology and the Internet, people have exposed to more and more digital images, therefore, how to search a desired image from a large number of digital images, has become a hot topic. Traditional text-based image retrieval, with some unavoidable shortcomings, has been unable to meet the growing demand. Therefore, content-based image retrieval technology comes into being.Content-based image retrieval is mainly based on some visual image features, such as color, texture, shape, etc., extract the image eigenvector, by calculating the similarity of eigenvectors to retrieve images. This paper studies the main feature of the image - the texture, analyzes the image texture based on spectral graph theory, extracts texture features and applies to the field of content-based image retrieval.First, the spectral graph theory is introduced into the field of image retrieval. The spectral graph theory in image segmentation has developed maturely in image segmentation, but has not been applied in image retrieval. In accordance with the gray distribution,the image is divided into N sub-images, and we can calculate the similarity of images by Normalized Cut between the sub-images to extract the image eigenvector.Then, construct an adaptive image texture analysis algorithm based on the spectral graph theory. According to the gray distribution of each image, using the image's energy, adjust image's classification adaptively, so make it correspond to each image comfortablely.Compared to non-adaptive approach to cut the sub-images, the adaptive algorithm reduces feature extraction errors and improve the retrieval accuracy.Finally, via a large number of experiments, compare the non-adaptive aigorithm to adaptive one, and compare the texture analysis algorithm based on the spectral graph to the traditional method of the texture image retrieval, it proves that the adaptive algorithm is superior to the traditional texture image retrieval algorithm and can achieve more satisfactory results.