Neural network based image deblurring

Abstract

In this paper, we propose a learning based technique for imagedeblurring using artificial neural networks. We model the original image as Markov Random field and the blurred image as degraded version of the original MRF. We do not make any prior assumptions for the blur kernel and develop the proposed algorithm by taking into account the space varying nature of the blur kernel. We re-formulate the image deblurring problem problem in terms of learning the mapping between original-MRF (original image) and degraded-MRF (blurred image), which is generally nonlinear. Instead of learning parameters of proposed MRF, a simple three layer neural network with backpropagation algorithm is used to learn the desired nonlinear mapping. Results of the experimentation on real data are presented.