Edge-preserving Regularized Image Restoration Based on Detail Distribution Estimation by Wavelet

The detail distribution of the real image estimated from a blurred image is important to the edge-preserving regularized image restoration.The method proposed in the paper combines wavelet details in several scales by seeking correlation operation.And the resultant image could reflect the edge distribution of the real image.Based on the Markov random field,proper potential function is chosen and the wavelet subband image is used to replace the gradient to construct the punitive function,so the number of regularization parameters and the training complexity are reduced.Finally,adaptive method is designed to assign the values to the parameters in the solution.Experimental results prove that the visual effect of the restoration result by this method is better than the common method using gradient,in particular better for processing the images in large degree of blurring.