Demosaicing and Super-resolution for Color Filter Array via Residual Image Reconstruction and Sparse Representation

Abstract

A novel approach of demosaicing and super-resolution for Color Filter Array (CFA) based on residual image reconstruction and sparse representation is proposed. Given an intermediate image produced by certain demosaicing and super-resolution, a residual image between a final reconstruction image and the intermediate image is reconstructed using sparse representation. Richer edges and details are found in the final reconstruction image. Specifically, a generic dictionary is learned from a large set of composite training data composed of intermediate data and residual data. The learned dictionary implies a mapping between the two data. A specific dictionary adaptive to the input CFA is learned thereafter. Using the adaptive dictionary, the sparse coefficients of intermediate data are computed and transformed to predict residual image. The residual image is added back into the intermediate image to obtain the final reconstruction image. Experimental results confirm the state-of-the-art performance in terms of PSNR and subjective visual perception.