Abstract
Image compression is an essential approach for decreasing the size in bytesof the image without deteriorating the quality of it. Typically, classicalgorithms are used but recently deep-learning has been successfully applied.In this work, is presented a deep super-resolution work-flow for imagecompression that maps low-resolution JPEG image to the high-resolution. Thepipeline consists of two components: first, an encoder-decoder neural networklearns how to transform the downsampling JPEG images to high resolution.Second, a combination between Generative Adversarial Networks (GANs) andreinforcement learning Actor-Critic (A3C) loss pushes the encoder-decoder toindirectly maximize High Peak Signal-to-Noise Ratio (PSNR). Although PSNR is afully differentiable metric, this work opens the doors to new solutions formaximizing non-differential metrics through an end-to-end approach betweenencoder-decoder networks and reinforcement learning policy gradient methods.