Fast Parameter Estimation in Image Restoration Based on Hierarchical Bayesian Framework

In this paper, an algorithm under Bayesian framework is proposed using the variational method to estimate the original image and parameters simultaneously and iteratively. The harmonic model and the gamma model are introduced as prior models for the original image and parameters separately. The harmonic model is a kind of smoothing model which is computationally efficient compared with nonlinear models, and it can preserve edges in the image better than most other smoothing models. The gamma can incorporate more prior knowledge for than the way that supposes the parameters as certain constants. The experimental results show the competitive performance of the proposed algorithm when used to estimate the parameters.