NON-LOCAL SIMILARITY DICTIONARY LEARNING BASED FACE SUPER-RESOLUTION

Face Super-Resolution (SR) is the process of producing a high-resolution face image from a set of low-resolution face images. Most existing dictionary learning based algorithms suffer a high degree of computational complexity and noise sensitivity. To solve this problem, we proposed a novel face SR method based on non-local similarity and multi-scale linear combination (NLS-MLC). Multi-scale linear combination consistency is proved under different resolutions. Experimental results show that the proposed SR method is more robust to noise and computationally efficient.