Research Of Texture Feature Extraction Algorithm Based On Regional Avery Binary Gray Level Difference Pattern

As an important characteristic for analyzing many types of images, texture is paidmore and more attention. However, because of characteristics lost, computationalcomplexity, high dimension and other problems, the existing texture feature extractionalgorithms don’t achieve the satisfactory extraction effect. As a result, it requires urgentsolutions currently how to extract texture feature which can describe the imageinformation more accurately. The research in this paper emphasizes how to enhance thevalidity of extracting texture feature, and the main works of this thesis are summarized asfollows.Firstly, through the research of texture structural analysis, this paper propose a newtexture primitive pattern--- regional average binary gray level difference, and this patterncan transform the original image to another pattern image with more distinct texture.According to this pattern image, a texture feature extraction algorithm is put forwardbased on regional average binary gray level difference co-occurrence matrix, combiningwith the texture statistical mind. The algorithm can accurately denote space relationshipbetween region pair of different texture patterns in the image by constructing regionalco-occurrence matrix about patter image and extracting second-order statistic parametersas the texture feature vectors. It cannot only capture subtle change of regional graytexture in the image, but also depict texture characteristics of the whole image.Secondly, the non-completedly tree-structure wavelet decomposition is brought tothe image of regional average binary gray level difference based on texture analysismethod of multiresloution, and we propose a texture feature extraction metahod ofregional average binary gray level difference based on tree-structure wavelet. Thismethod makes the extracting texture feature have multiresolution by combining structuralanalysis with spectral analysis, which can achieve the goal of analyzing image withtexture multiresolution and gain the detail information of image texture in differentmultiresolutions.Finally, an experiment of image matching is designed to compare with the traditional texture feature extraction algorithms, and it validates the correctness of thealgorithm of extracting texture feature presented in this paper througt analyzing thematching accurancy of these methods.