Learning adaptive local distance metric for face hallucination

摘要: In this paper, we propose a novel method for face hallucination by learning new distance metric in the low-resolution (LR) patch space (source space). Local patch-based methods usually assume that two manifolds formed LR and high-resolution (HR) image patches have similar local geometry. However, assumption does not hold well practice. Motivated machine learning, to learn source space, under supervision of true geometry target (HR The learned gives more freedom presentation thus geometries turn be consistent. Experiments conducted on datasets demonstrate proposed is superior state-of-the-art super-resolution (SR) methods.