The Research On Texture-Based Image Retrieval

With the extensive application of computer technologies and network technology, a large number of daily digital images will exist in the press and publishing, health care, building design and other industries. How to effectively analyze, store and retrieve these images is an imperative problem. The content-based image retrieval technology can effectively resolve the problem and it has also become a hotspot in study. Texture is one of the main characteristics of images as well as an important means based on content-based image retrieval system, therefore, the subject is about image retrieval research based on the texture features.We firstly review and compare different methods of texture feature extraction and texture analysis methods, and categorize some common similarity measures, pointing out strengths and limitations of each method. Our research focus on the extraction of texture features and the similarity measure of texture features.The work done in the aspects of texture features and image retrieval is as follows:1 Histogram of the moment is chosen to extract texture feature because histogram of the moment directly reflects the aspects in pixels changing with gray value. Though other methods have their advantages, pixel gray-scale statistical features of gray level co-occurrence matrix do not establish corresponding relation with the visual identification of human beings in the identification of texture features, the texture features of Gabor wavelet function is not stable in turning.2 The nature of image texture features are analyzed by using Tamura texture features for texture analysis. Although other methods have their advantages, the run-length statistical analysis does not fully reflect the rules of the overall pattern of image and the overall pixel,Search Results of Fourier transform and wavelet transform texture analysis have a certain degree of distortion.3 As a result of the texture features inherent characteristics of regionality and locality, Pure form method identify most similar images with the sample image texture features by calculating the distance between pixels and determining the direction and doing a series of linear transformation.Distance function similarity measure: Simple calculation, but limited conditions are too strict. Hausdorff distance: Matching is not sensitive to the deformation, but sensitive to noise. Mahalanobis distance only requires that relevance and different weights should be provided between each component of eigenvector.In this paper, image retrieval system based on texture feature is designed and implemented. The system could search image base according to given sample images. And the matching images are ordered by descending according to the degree of similarity values with sample image. The experimental results show that experimental system has better retrieval performance.