A technique based on regularization method and restores image to close-to-optimal is proposed in this paper. The less the energy of the regularized residue, the better the image restoration. Based on this idea, wavelet transform is employed to choose regularization operator qualitatively, and stochastic theory is used to calculate the expectation of the energy, by minimizing the expectation to determine regularization parameter. Qualitative analysis concludes that the regularization operator should be low-stop and high-pass, and the experimental results show that the performances of this method are better than the traditional methods and yields steadily close-to-optimal restoration.