Super-Resolution for Face Recognition Improvement

Super-Resolution is a class of Computer Vision techniques that solve a problem of image up-scaling without loss of quality. Usually, they used for improving images in terms of human perception of the beauty. Face recognition is another problem in Computer Vision domain. Solutions in this field help to discriminate and identify people using images of their faces. This paper provides a number of experiments on the usage of Super-Resolution for facial images up-scaling for further better recognition of them. This research shows how basic up-sampling approaches based on neural networks, usage of perceptual loss function and application of Generative Adversarial Neural Networks influence on the quality of solving the defined problem. This paper presents different ways of a deep facial feature extracting and advantages of using of neural networks trained for Face Recognition problem solving for perceptual loss over standard approaches to extract this features.