JPEG Artifact Reduction Based on Deformable Offset Gating Network Controlled by a Variational Autoencoder

For the reduction of JPEG compression artifacts, there have been many methods using deep neural networks. Most them use quality factor (QF) as prior knowledge in designing and training However, since images we get from Internet are often recompressed, given QF is not so informative or misleading. Also, early works validated their on low QFs less than 50, while recent smartphones high larger equal to 90. In this paper, propose a new artifacts network considering above-stated problems. Specifically, extract information input image itself instead provided header file, variational autoencoder (VAE) regard its latent vector information. artifact network, let change flexibly according by employing deformable offset gating (DOG) network. The VAE merged our overall dubbed DOG-VAE, where used adjust DOG quality. DOG-VAE trained end-to-end with range [10,90]. Extensive experiments validate that method achieves comparable results state-of-the-art for monochrome better color images. Our codes available at https://github.com/yunjh410/DOGNet.