Parametric model-adaptive image restoration
We have developed two parametric model-adaptive image restoration methods. The first one is the model-adaptive maximum likelihood method based on the generalized Gaussian noise model. Improved performance over many widely used methods is obtained by adapting the optimization criteria to the observed data. The second method is the generalized Gaussian Markov random field (GGMRF) image prior method. Image structure is modeled by adjusting a shape parameter of the GGMRF. The GGMRF model can lead to better results than that of the popular Gaussian Markov random field (GMRF) model. In addition, results from using the GGMRF model are more pleasing to human visual perception. Parameter estimation methods and fast iterative algorithms are also presented to complement the development of these two methods.