Densely Residual Laplacian Super-Resolution

28 Jun 2019  ·  Saeed Anwar, Nick Barnes ·

Super-Resolution convolutional neural networks have recently demonstrated high-quality restoration for single images. However, existing algorithms often require very deep architectures and long training times. Furthermore, current convolutional neural networks for super-resolution are unable to exploit features at multiple scales and weigh them equally, limiting their learning capability. In this exposition, we present a compact and accurate super-resolution algorithm namely, Densely Residual Laplacian Network (DRLN). The proposed network employs cascading residual on the residual structure to allow the flow of low-frequency information to focus on learning high and mid-level features. In addition, deep supervision is achieved via the densely concatenated residual blocks settings, which also helps in learning from high-level complex features. Moreover, we propose Laplacian attention to model the crucial features to learn the inter and intra-level dependencies between the feature maps. Furthermore, comprehensive quantitative and qualitative evaluations on low-resolution, noisy low-resolution, and real historical image benchmark datasets illustrate that our DRLN algorithm performs favorably against the state-of-the-art methods visually and accurately.

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Image Super-Resolution BSD100 - 2x upscaling DRLN+ PSNR 32.47 # 2
SSIM 0.9032 # 1
Image Super-Resolution BSD100 - 3x upscaling DRLN+ PSNR 29.4 # 2
SSIM 0.8125 # 1
Image Super-Resolution BSD100 - 4x upscaling DRLN+ PSNR 27.87 # 2
SSIM 0.7453 # 9
Image Super-Resolution BSD100 - 8x upscaling DRLN+ PSNR 25.06 # 1
SSIM 0.607 # 2
Image Super-Resolution Manga109 - 2x upscaling DRLN+ PSNR 39.75 # 1
SSIM 0.9792 # 1
Image Super-Resolution Manga109 - 3x upscaling DRLN+ PSNR 34.94 # 1
SSIM 0.9518 # 1
Image Super-Resolution Manga109 - 4x upscaling DRLN+ PSNR 31.78 # 3
SSIM 0.9211 # 3
Image Super-Resolution Manga109 - 8x upscaling DRLN+ PSNR 25.55 # 2
SSIM 0.8087 # 2
Image Super-Resolution Set14 - 2x upscaling DRLN+ PSNR 34.43 # 2
SSIM 0.9247 # 2
Image Super-Resolution Set14 - 3x upscaling DRLN+ PSNR 30.8 # 2
SSIM 0.8498 # 1
Image Super-Resolution Set14 - 4x upscaling DRLN+ PSNR 29.02 # 5
SSIM 0.7914 # 10
Image Super-Resolution Set14 - 8x upscaling DRLN+ PSNR 25.4 # 2
SSIM 0.6547 # 3
Image Super-Resolution Set5 - 2x upscaling DRLN+ PSNR 38.34 # 2
SSIM 0.9619 # 2
Image Super-Resolution Set5 - 3x upscaling DRLN+ PSNR 34.86 # 2
SSIM 0.9307 # 1
Image Super-Resolution Set5 - 4x upscaling DRLN+ PSNR 32.74 # 4
SSIM 0.9013 # 8
Image Super-Resolution Set5 - 8x upscaling DRLN+ PSNR 27.46 # 4
SSIM 0.7916 # 4
Image Super-Resolution Urban100 - 2x upscaling DRLN+ PSNR 33.54 # 2
SSIM 0.9402 # 2
Image Super-Resolution Urban100 - 3x upscaling DRLN+ PSNR 29.37 # 3
SSIM 0.8746 # 1
Image Super-Resolution Urban100 - 4x upscaling DRLN+ PSNR 27.14 # 7
SSIM 0.8149 # 9
Image Super-Resolution Urban100 - 8x upscaling DRLN+ PSNR 23.24 # 1
SSIM 0.6523 # 1

Methods


No methods listed for this paper. Add relevant methods here