Parameter selections for Tikhonov regularization image restoration

The model of image degradation due to atmospheric turbulence can be decomposed into two one-dimensional normal degenerations in horizontal and vertical directions successively. The recovery is an inverse process of degeneration. Each column of blurred image was restored by one-dimensional regularization method, then each row of restored image in vertical direction was recovered with same method. The regularization parameter was selected with the L-curve criterion, GCV and UPRE method respectively, when the degenerated image was restored in every column, then in every row, and different recovery results were obtained with different parameter selections. Simulation results show that if the blurred image has high SNR, three types of regularization parameter selection methods reached similar accuracy in image restoration, the GCV method which don´t need a priori variance of the noise is more stable and effective than other two methods.