Super Resolution Image Reconstruction by Granular Computing with L1-norm

Abstract

According to the higher computational complexity during the training process of sparse representation, the centers of granular computing (GrC) with L1-norm are regarded as the bases of sparse representation and used to reconstruct the super-resolution image of input image. Firstly, the granule is represented as the shape of hyperdiamond by the L1-norm in N-dimensional space. Secondly, the join operation between two hyperdiamond granules is designed to transform the microcosmic world into the macroscopic world. Thirdly, the threshold r of granularity is used to control the join process. The centers of granules are regarded as the approximate bases to reconstruct the super-resolution (SR) image of the low-resolution (LR) image. Experimental results show that the SR image reconstruction by GrC with L1-norm reduced the root mean square error (RMSE) between the SR image and the original image compared with the bicubic interpolation and sparse representation.