Ehrlich Max ; Davis Larry ; Lim Ser-Nam ; Shrivastava Abhinav

Abstract

The JPEG image compression algorithm is the most popular method of imagecompression because of its ability for large compression ratios. However, toachieve such high compression, information is lost. For aggressive quantizationsettings, this leads to a noticeable reduction in image quality. Artifactcorrection has been studied in the context of deep neural networks for sometime, but the current state-of-the-art methods require a different model to betrained for each quality setting, greatly limiting their practical application.We solve this problem by creating a novel architecture which is parameterizedby the JPEG files quantization matrix. This allows our single model to achievestate-of-the-art performance over models trained for specific quality settings.