An Augmented Lagrangian Method for the Patch-based Gaussian Mixture Model In Image Deblurring

Since the image deblurring problem is ill-conditioned, a good regularization term can improve the quality of the deblurred result greatly. The expected patch log likelihood (EPLL) is a patch-based regularization prior working with small image patches, it has been shown to be effective and achieves good performance for image deblurring. However, in the EPLL method, it needs the auxiliary parameter becomes large which results to numerical difficulty. To avoid such a problem, we adapt the augmented Lagrangian method into the EPLL image deblurring algorithm. Experimental results show that the proposed deblurring method makes an improvement in image quality and outperforms the existing image deblurring algorithms.