Robust Face Hallucination via Locality-Constrained Nuclear Norm Regularized Regression

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