Spatio-temporal Complex Wavelet Based Dynamic Texture Classification

Dynamic texture,as a special visual texture,has become an important part in the field o f the texture research,and has been widely used in many fields.The so-called dynamic textu re classification is what utilizes the texture attributes to identify the area or the type of the object.With the development the computer technology,dynamic texture classific ation is becoming a frontier research with the high scientific value.The limitation of the traditional wavelet transform impedes the application of wave transform is the study of the dynamic texture,because the traditional wavelet transform has limitations.In this paper the Space-Time Complex Wavelet with good directional selectivity and translation invariance instead of the wavelet can achieve the good combination of the“movement” of the texture characteristics and the “appearance” of the texture characteristics.Thereby,the classification of the dynamic texture can be completed.The main work are as follows:1.The dynamic texture classification based on the generalized Gussian distributions.The generalized Gaussian distribution of the Spatio-temporal Complex Wavelet transform coefficients is utilized to create the statistical model for random variable,it can overcome the non-Gaussian of the dynamic texture sub-block coefficient.And then the maximum likelihood method is utilized to estimate the parameter of the model and to obtain the required dynamic texture characteristic.Next,a new texture characteristic based on the finite complex generalized Gaussian distribution parameters is proposed.Finally,the classification experiment is completed by the k-nearest neighbor classifier.2.The dynamic texture classification based on the extreme value distribution.Firstly,the concept of the extreme value distribution is described.And the extreme value theory is integrated into the wavelet domain,the space-time complex wavelet transform is utilized to divide the dynamic texture into sub-block.Then,the two different types of the extreme value distribution,I type Gumbel extreme value distribution and the generalized Pareto distribution,is used to model the median of the dynamic texture sub-block,and the parameters of the model are estimated respectively and the estimated parameter is the eigenvalues of the dynamic texture classification.Finally the classification performance of the novel eigenvaluesare compared with the classification performance of the eigenvalues based on the energy to verify the effectiveness of the proposed approach.