Quantization Guided JPEG Artifact Correction

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

PDF Abstract ECCV 2020 PDF ECCV 2020 Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
JPEG Artifact Correction BSDS500 (Quality 10 Color) QGAC PSNR 27.69 # 2
PSNR-B 27.36 # 2
SSIM 0.810 # 1
JPEG Artifact Correction BSDS500 (Quality 10 Grayscale) QGAC PSNR 29.54 # 2
PSNR-B 29.04 # 2
SSIM 0.833 # 1
JPEG Artifact Correction BSDS500 (Quality 20 Color) QGAC PSNR 29.89 # 2
PSNR-B 29.29 # 2
SSIM 0.876 # 1
JPEG Artifact Correction BSDS500 (Quality 20 Grayscale) QGAC PSNR 31.79 # 2
PSNR-B 30.96 # 2
SSIM 0.894 # 1
JPEG Artifact Correction BSDS500 (Quality 30 Color) QGAC PSNR 31.15 # 2
PSNR-B 30.37 # 2
SSIM 0.903 # 1
JPEG Artifact Correction BSDS500 (Quality 30 Grayscale) QGAC PSNR 33.12 # 2
PSNR-B 32.42 # 1
SSIM 0.907 # 2
JPEG Artifact Correction Classic5 (Quality 10 Grayscale) QGAC PSNR 29.84 # 4
PSNR-B 29.43 # 2
SSIM 0.837 # 1
JPEG Artifact Correction Classic5 (Quality 20 Grayscale) QGAC PSNR 31.98 # 4
PSNR-B 31.37 # 2
SSIM 0.885 # 1
JPEG Artifact Correction Classic5 (Quality 30 Grayscale) QGAC PSNR 33.22 # 4
PSNR-B 32.42 # 2
SSIM 0.907 # 1
JPEG Artifact Correction ICB (Quality 10 Color) QGAC PSNR 32.11 # 2
PSNR-B 32.47 # 1
SSIM 0.815 # 1
JPEG Artifact Correction ICB (Quality 10 Grayscale) QGAC PSNR 34.73 # 1
PSNR-B 34.58 # 1
SSIM 0.896 # 1
JPEG Artifact Correction ICB (Quality 20 Color) QGAC PSNR 34.23 # 2
PSNR-B 34.67 # 1
SSIM 0.845 # 1
JPEG Artifact Correction ICB (Quality 20 Grayscale) QGAC PSNR 37.12 # 1
PSNR-B 36.88 # 1
SSIM 0.924 # 1
JPEG Artifact Correction ICB (Quality 30 Color) QGAC PSNR 35.20 # 2
PSNR-B 35.67 # 1
SSIM 0.860 # 1
JPEG Artifact Correction ICB (Quality 30 Grayscale) QGAC PSNR 38.43 # 1
JPEG Artifact Correction LIVE1 (Quality 10 Color) QGAC PSNR 27.65 # 2
PSNR-B 27.40 # 4
SSIM 0.819 # 1
JPEG Artifact Correction Live1 (Quality 10 Grayscale) QGAC PSNR 29.53 # 5
PSNR-B 29.15 # 7
SSIM 0.840 # 1
JPEG Artifact Correction LIVE1 (Quality 20 Color) QGAC PSNR 29.92 # 3
PSNR-B 29.51 # 7
SSIM 0.882 # 1
JPEG Artifact Correction LIVE1 (Quality 20 Grayscale) QGAC PSNR 31.86 # 5
PSNR-B 31.27 # 8
SSIM 0.901 # 1
JPEG Artifact Correction LIVE1 (Quality 30 Color) QGAC PSNR 31.21 # 2
PSNR-B 30.71 # 2
SSIM 0.908 # 1
JPEG Artifact Correction LIVE1 (Quality 30 Grayscale) QGAC PSNR 33.23 # 4
PSNR-B 32.50 # 2
SSIM 0.925 # 1

Methods


No methods listed for this paper. Add relevant methods here