Complex Wavelet-Based Texture Classification Using Generalized Gaussian Density

Texture is the performance of image content of the most significant one ofthe basic visual features, is the image analysis and image understanding a veryimportant source of information. Describes the image or images, the correspond-ing features of the surface properties, Contains the ordered surface structure oforganizations, important information and their links with the surrounding envi-ronment [1]. Image texture analysis is to understand, analyze and identify theimportant research. Texture has been an important research field of image pro-cessing elements.Dual-tree complex wavelet transform not only maintained good traditionalwavelet transform analysis of time-frequency localization ability, also has nearlyLike translation invariance and good direction analysis ability This article usesdual tree complex wavelet transform, texture image, and get dual tree complexwavelet coefficients. On this basis, this paper applied a texture image transformdomain extraction of statistical features. Histogram of wavelet coefficients due toobey generalized Gaussian distribution [28], This article proposes the applicationof generalized Gaussian distribution of the obtained dual tree complex wavelet co-efficients of the model of the histogram for the maximum likelihood estimate. Willbe the parameters as texture features. And this paper, support vector machineelements on the feature space for training and testing, Verified by experiment thevalidity of this method, Also compared with some of the existing methods showthat the use of texture image classification method has higher recognition rate.The first chapter is an introduction, a general introduction to the significanceof the research of this article, current research methods.Chapter II briefly introduces the multi-scale analysis and Mallat algorithm,And on this basis, introduced dual-tree complex wavelet basic concepts and prin-ciples.Chapter III briefly introduces the VC theory, optimization theory, the nuclear function theory, and on this basis, introducing support vector machine theory.Chapter IV is the focus of this article. Mainly for texture image binary treecomplex wavelet transform, complex wavelet coefficients obtained from applicationof the modulus value Generalized Gaussian maximum likelihood estimation. Theparameters obtained as a texture image features and support vector machines,texture image classification Proved by experiment method. Also compared withsome of the existing methods show that the texture feature extraction accuracy.This texture image classification method can be summarized as the followingsteps:Step 1 The texture image for dual-tree complex wavelet transform (decom-position level for this article 4), calculated by each layer of the six high-frequencysub-band, Corresponds to each frequency sub-band and imaginary parts really.Step 2 For each frequency sub-band complex wavelet coefficients modelwith the generalized Gaussian maximum likelihood estimation, the parametersobtained as eigenvalues In order to eliminate dimensional characteristics of thedi?erent impact on the following classi?cation, therefore, usually linear scaling ofdata to the [-1,1] range.Step 3 On the feature set, the application of support vector machine classi-fication on the characteristics and to achieve the right texture image classification.Numerical experiments are presented method and K-means clusteringmethod, based on wavelet transform domain methods Literature [27] in themethod of comparison, verify the classification of texture images used in thevalidity and accuracy.