Image compression based on a family of stochastic models

Abstract

In this paper, we propose a family of stochastic models for image compression, where images are assumed to be Gaussian Markov random field. This model is based on stationary full range autoregressive (FRAR) process. The parameters of the model are estimated with the Monte-Carlo integration technique based on Bayesian approach. The advantage of the proposed model is that it helps to estimate the finite number of parameters for the infinite number of orders. We use arithmetic coding to store seed values and parameters of the model as it gives furthermore compression. We also studied the use of Metropolis–Hastings algorithm to update the parameters, through which some image contents such as untexturedness are captured. Different types–both textured and untextured images–are used for experiment to illustrate the efficiency of the proposed model and the results are encouraging.