Dynamic Textures Segmentation Based on Markov Random Field

In recent dynamic texture segmentation, It remains a problem in pattern recognition, image and computer vision. Dynamic texture segmentation problem can be solved by two ways: non-parametric classification and model fitting. In this paper, we use MRF in unsupervised dynamic texture segmentation. We present a novel MRF parameter estimation method based on MCMC (Markov chain Monter Carlo) approach. The MCMC approach is formulate to allow the sampling of the parameters from the posterior distribution of the dynamic texture. The experiments show that the method gives a good estimation result and it is suitable to segment dynamic texture.