Robust face super-resolution via iterative sparsity and locality-constrained representation

The performance of image super-resolution (SR) process is highly affected by impulse noise hence, a novel iterative sparsity and locality-constrained representation (ISLcR) based face super-resolution model is proposed here. The proposed model computes data fidelity in high-resolution (HR) and low-resolution (LR) face spaces aiming to compensate the lost information in LR space from the HR space. For this purpose, supporting HR face is computed using the proposed ISLcR model. Further, reconstruction residual of both data fidelity is handled by the locality with sparsity regularization term. The use of both types of data fidelity and locality with sparsity regularization help in reduction of noise, generate a more discriminable outcome, and makes the process computationally viable. The superiority of the proposed ISLcR based face SR model over existing state-of-the-art methods has been established by conducting the experiments on public standard human face datasets and the images from locally recorded surveillance video. The experimental results indicate that the proposed model has outshined all the existing methods.