A novel sparse representation based framework for face image super-resolution

Abstract

In this paper, we present a new face image super-resolution framework using the sparse representation (SR). Firstly, a mapping function between the embedding geometries in respective image space is estimated from a training set. For super-resolution, we first seek a sparse representation for each low-resolution (LR) input, and then the representation coefficients are mapped to generate the corresponding representation coefficients in high-resolution (HR) space. Finally, the mapped coefficients are used to reconstruct the initial estimation of the target HR image. To obtain the HR images with higher fidelity, the maximum a posteriori (MAP) formulation is introduced. The effectiveness of the proposed method is evaluated through the experiments on the benchmark face database, and the experimental results demonstrate that the proposed method can achieve competitive performance compared with other state-of-the-art methods.