This paper is devoted to the combination of several prior models in Bayesian image restoration and increasingly wide utilization in astronomical images. Bayesian methods introduce image models using prior knowledge and address the ill-posed problem in the registration parameter estimation. Employing a variational Bayesian analysis, we obtain a unique approximating distribution based on the observations that decreases the Kullback Leibler distance for more optimal posterior distribution. The estimated results on astronomical images experimentally provide higher quality and better restoration performance.