Dong Chao Deng Yubin Loy Chen Change Tang Xiaoou

Abstract

Lossy compression introduces complex compression artifacts, particularly theblocking artifacts, ringing effects and blurring. Existing algorithms eitherfocus on removing blocking artifacts and produce blurred output, or restoressharpened images that are accompanied with ringing effects. Inspired by thedeep convolutional networks (DCN) on super-resolution, we formulate a compactand efficient network for seamless attenuation of different compressionartifacts. We also demonstrate that a deeper model can be effectively trainedwith the features learned in a shallow network. Following a similar "easy tohard" idea, we systematically investigate several practical transfer settingsand show the effectiveness of transfer learning in low-level vision problems.Our method shows superior performance than the state-of-the-arts both on thebenchmark datasets and the real-world use case (i.e. Twitter). In addition, weshow that our method can be applied as pre-processing to facilitate otherlow-level vision routines when they take compressed images as input.