Robust face hallucination via locality-constrained nuclear norm regularized regression

Abstract

The performance of traditional face recognition approaches is sharply reduced when facing a low-resolution (LR) probe face image. Various face hallucination methods have been proposed in the past decade to obtain much more facial details. The basic idea of face hallucination is to desire a high-resolution (HR) face image from an observed LR one with the help of a set of training examples. In this paper, we propose a locality-constrained nuclear norm regularized regression (LCNNR) model for face hallucination task and use the alternating direction method of multipliers to solve it. LCNNR attempts to directly use the image matrix to compute the representation coefficients to maintain the essential structural information. Moreover, a locality constraint is also enforced to preserve the locality and the sparsity simultaneously. Experiments carried out on the benchmark FEI face database show that LCNNR outperforms some state-of-the-art algorithms.