Graphical Models and Contourlet Image Denoising

Recently Conditional Random Fields (CRF), have presented to be a promising model for statistical inference in the vast literature of computer vision.CRFs are flexible to model long-range statistics by directly modelling the posterior distribution without any explicit prior model describing the observation. In this paper we propose a CRF-based method to denoise the contourlet coefficients and to examine effectiveness of this new tool in comparison with the state-of-the-art denoisers such as hidden Markov trees. Our interest is to implement denoising in the contourlet domain with respect to the coefficient correlations of inter-scale and also intra-scale, by defining a graphical structure within this domain.

Keywords:

Conditional Random Fields , Contourlet , Feature Function , Tree Structure , Denoising