Facial Image Super Resolution and Feature Reconstruction using SRGANs with VGG-19-based Adaptive Loss Function

Image reconstruction and super resolution has various applications. Several deep learning techniques are being employed to constantly improve this space. The aim of this experiment is to showcase a unique deep learning approach to try and super resolve human faces from low resolution images. The experiment makes use of a machine learning framework designed to improve image quality called Super Resolution Generative Adversarial Neural (SRGANs) with a loss function based on the features accumulated from multiple layers of a trained Convolutional Neural Network named Visual Geometry Group-19 (VGG-19). The model super resolves lower quality image input and gives out image output of a superior quality