ABSTRACT
Image Super-Resolution (SR) is an ill-posed problem with an objective of constructing High Resolution (HR) image from given Low Resolution (LR) image. The existing popular SR approaches include deep neural networks which require large dataset for training and also they are computationally expensive. In this paper, a novel algorithm to achieve SR which utilizes examples taken from LR itself for training have been proposed. The proposed algorithm fastens the training process via selectively fetched examples (patches) from given LR image. For the training purpose, Simultaneous Codeword Optimization (SIMCO) based dictionary learning algorithm is used along-with Least Absolute Shrinkage and Selection Operator (LASSO) for sparse representation. The trained dictionary is further utilized for constructing an HR image. Experimental results show that the proposed algorithm outperforms in terms of computational complexity and quantitative measures also with respect to deep learning algorithms.
