應用生成對抗網絡與用圖像品質評估的變換器

Abstract

none

Image Quality Enhancement is the process in which images can be improved for better human interpretation of its contents. Image enhancement is usually done based on certain parameters specified when formulating the problem. Generative Adversarial Networks (GAN), on the other hand, can create new images based only on characteristics it finds on the training set, without specifying those characteristics. We utilize three variations of the GAN architecture, Cycle GAN, Conditional GAN and EnlightenGAN, to implement different solutions to generate images with an increased image quality on existing datasets. Our goal is to demonstrate that Transformer for Image Quality Assessment, an image quality evaluation framework, can give a frame of reference for the performance of these GANs, and that those GANs can increase the quality of images after being trained on an original and enhanced group.