Image Restoration: Use of AR and IGMRF Priors

N. Khandelwal, N. Gupta, and M.V. Joshi (India)


Autoregressive Parameters, IGMRF parameters, gradient descent approach, Regularization term.


In this paper, we propose image restoration using two dif ferent priors. The degraded image to be restored is modeled as Gaussian blurred and noisy version of its original ver sion. Since this is an ill-posed inverse problem, we need proper regularization. Towards this end we first propose an approach based on homogenous autoregressive (AR) prior. We next consider the Inhomogeneous Gaussian Markov Random Field (IGMRF) prior. The AR and the IGMRF parameters are estimated using the initial estimated image obtained by Wiener filtering. A suitable cost function is formed for each of the approach which is then minimized using simple gradient descent optimization method. Sim ulation results for both gray scale and colour images are shown. Also the out of focus blur real images and videos captured from a real camera and the atmospheric turbu lence blur real images taken from internet are restored us ing the proposed algorithms.

Important Links:

Go Back