A Novel Technique for Image Super Resolution Based on Sparse Representations and Compact Entity Extraction

Abstract

A novel method of image super resolution using sparse representation has been discussed in this paper. The main purpose is to acquire the super-resolved image from the down scaled and blurred images. With the small number of elements from a huge set of vectors, sparse signal model approximates signals and this large dataset is called a dictionary. For construction of high and low-resolution dictionaries from the condensed atoms extracted from the training image patches, the Orthogonal Matching Pursuit approach has been used. The blurred and down-scaled version of the image is super resolved using the above-mentioned dictionaries. The outcomes are compared both instinctively by the visual assessment of the resulting super-resolve images by means of the proposed scheme and the bi-cubic interpolation method, and by comparing the Peak Signal-to-Noise Ratio (PSNR) obtained by the two approaches. Both the comparison metrics, i.e. visual quality of acquired super resolved images and PSNR measures show that the proposed approach is superior to the existing state of the art Bi-Cubic interpolation.