Simultaneous Denoising and Deblurring by Non-Convex Regularization: Alternating Minimization Framework

Abstract

In this paper, we have addressed the problem associated with simultaneous denoising and deblurring of a degraded image. We have implemented Discontinuity Adaptive Markov Random Field (DAMRF) algorithm along with a nonlocal based filtering technique. The image under consideration is degraded by Additive White Gaussian Noise (AWGN) and blur. We implemented a space-invariant blind deconvolution to deblur the image using DAMRF regularization frame work. Latent focused image along with the blur map sigma of the space-invariant point spread function (PSF) is estimated by regularization based optimization techniques. The problem is critically ill-posed as well as inverse hence it has several solutions. We exploit a recently proposed method with a MAP-DAMRF energy minimization to estimate latent image followed by a blur map sigma. Statistical analyses and derivations are carried out on effective priors, by utilizing the methods of non-local neighborhood pixels information (for denoising) and local neighbor-hood pixels information (for deblurring). Objective of our proposed work about alternating energy minimization is successfully optimized by implementing the gradient descent method. We showed that the performance metric of DAMRF prior is better than Gauss-Markov random field (GMRF) prior. Edges, fine details and structure information of 2D image generated by the DAMRF regularizer is better than few state-of-the-art work. Comparability of both PSNR and SSIM evaluation of our work with presented.