Dynamic Textures Segmentation Based On Markov Random Field And Non-sampling Wavelet Transform

The dynamic texture is texture image sequences with time correlation and it hasrepetitive in space. There are many aspects of the research about dynamic texture. Includingdynamic texture classification, segmentation and synthesis. The dynamic texture exists inmany aspects of people’s lives, and therefore has a very wide range of applications.Dynamic texture segmentation is a process that relabel several regions of the mutuallyoverlapping in the texture image sequence, and the regions have uniform texture. Markovrandom field model can effectively describe the features of dynamic texture. So, This articledeeply studies the dynamic texture segmentation algorithm in the view of Markov RandomField model. And we consider the multi-resolution, forming the markov random field modelin view of no-sampling wavelet transform. The main work of my paper as follows:1. Dynamic texture segmentation in view of MRF model. According to the using ofneighborhood systems and the energy function in the label field to describe the characteristicsof the dynamic texture. Respectively, using the label field obeying the Gibbs distribution, andthe observed field obeying the Gaussian distribution to describe the dynamic texture. UsingMCMC method to estimate parameters of the observed field, we get the conformation ofdynamic texture segmentation algorithm in the view of MRF model. Maximum a posteriori(MAP) is the standard of dynamic texture segmentation. The simulation results show theeffectiveness of MRF model.2. Dynamic texture segmentation based on non-sampling wavelet transform and Markovrandom field model. According to the translational invariance of non-sampling wavelettransform, non-sampling wavelet is more suitable for dynamic texture segmentation. Thedynamic texture segmentation algorithm in view of non-sampling wavelet transform isbrought forward in this paper. The algorithm considers time-consuming of the MCMC methodfor parameter estimation, proposing applicating improved MCMC method to estimateparameter. Reducing the computational complexity of dynamic texture segmentation. At thesame time, the algorithms considers the relation in the wavelet coefficients of each same scalesub-band, building the NSMRF model. The simulation experiment shows the superiority ofNSMRF model for dynamic texture segmentation.