Abstract:

Mixture proportions of the existing work don't have clear probability vector models, and some models cannot solve the iterative convergence problem of Markov chain Monte Carlo (MCMC). According to the Gaussian mixture models with spatial smoothing based on constraint, we present a new Bayesian model which can be applied in image segmentation. We use the probability density model of the latent Dirichlet allocation (LDA)  and the latent Dirichlet parameters of Gauss-Markov random field (MRF) to achieve parameter smoothing process. The proposed model has two advantages: 1) the model for the spatial smoothing constraint defines the probability vector model proportion; 2) we use the maximum a-posteriori (MAP) and the expectation maximization (EM) to achieve the update of closed parameters. Experimental results show that the proposed model has better image segmentation performance than the GMM method, and it has been successfully applied in the image segmentation of natural images and natural artistic images with noise.

Key words: Bayesian model, latent Dirichlet allocation(LDA), Gaussian mixture model, image segmentation(GMM), expectation maximization(EM) method