Super-Resolution Image Reconstruction Using MRF Model

Abstract

Super Resolution of an image is one of the image processing methods that helps us in estimating the high-resolution (HR) images from a set of low-resolution (LR) images or from a single low resolution image. In most of the practical applications, the resolution outputs produced by the imaging sensors are indigent. When improvement of resolution cannot be made by replacing other sensors, either due to cost or physical hardware limitations, it is necessary for us to rely on methods of Super Resolution. Even in cases when high resolution imaging sensors of high cost are available, Super Resolution methods act as an inexpensive alternative.
This thesis introduces an enhanced image super resolution reconstruction method based on Markov random field (MRF) model. Prior to that image degradation is removed using simulated annealing algorithm and MRF model. Then the reconstruction step is achieved using Super Resolution method. This work incorporates an efficient training set. The most appropriate training set is utilized to find the similar patches based on the associated local activity of the image patch to improve the quality of reconstruction and to alleviate the computational overload. Secondly, MRF model is formulated to find the correspondence between low resolution patches of the test image and the candidate high resolution patches of the training data set. Many Super Resolution methods have been proposed over the years which have its own advantages and disadvantages. However, of all the methods learning based Super Resolution using MRF model is the most efficient method. A guiding insight underlying most of the work on Markov Random Field models in image processing is that the information contained in the local, physical structure of images is sufficient to obtain a good, global image representation. Experimental results demonstrate the outperforming
Characteristics of the proposed method in terms of reconstruction quality and computational time than several existing state of the art methods.