Residual Dense Network for Image Restoration

25 Dec 2018  ·  Yulun Zhang, Yapeng Tian, Yu Kong, Bineng Zhong, Yun Fu ·

Convolutional neural network has recently achieved great success for image restoration (IR) and also offered hierarchical features. However, most deep CNN based IR models do not make full use of the hierarchical features from the original low-quality images, thereby achieving relatively-low performance... In this paper, we propose a novel residual dense network (RDN) to address this problem in IR. We fully exploit the hierarchical features from all the convolutional layers. Specifically, we propose residual dense block (RDB) to extract abundant local features via densely connected convolutional layers. RDB further allows direct connections from the state of preceding RDB to all the layers of current RDB, leading to a contiguous memory mechanism. To adaptively learn more effective features from preceding and current local features and stabilize the training of wider network, we proposed local feature fusion in RDB. After fully obtaining dense local features, we use global feature fusion to jointly and adaptively learn global hierarchical features in a holistic way. We demonstrate the effectiveness of RDN with several representative IR applications, single image super-resolution, Gaussian image denoising, image compression artifact reduction, and image deblurring. Experiments on benchmark and real-world datasets show that our RDN achieves favorable performance against state-of-the-art methods for each IR task quantitatively and visually. read more

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Grayscale Image Denoising BSD68 sigma10 Residual Dense Network + PSNR 34.01 # 1
Color Image Denoising BSD68 sigma10 Residual Dense Network + PSNR 36.49 # 1
Grayscale Image Denoising BSD68 sigma30 Residual Dense Network + PSNR 28.58 # 1
Color Image Denoising BSD68 sigma30 Residual Dense Network + PSNR 30.7 # 1
Grayscale Image Denoising BSD68 sigma50 Residual Dense Network + PSNR 26.43 # 3
Color Image Denoising BSD68 sigma70 Residual Dense Network + PSNR 26.88 # 1
Grayscale Image Denoising BSD68 sigma70 Residual Dense Network + PSNR 25.12 # 2
JPEG Artifact Correction Classic5 (Quality 10 Grayscale) Residual Dense Network + PSNR 30.03 # 1
SSIM 0.8194 # 2
JPEG Artifact Correction Classic5 (Quality 20 Grayscale) Residual Dense Network + PSNR 32.19 # 1
SSIM 0.8704 # 2
JPEG Artifact Correction Classic5 (Quality 30 Grayscale) Residual Dense Network + PSNR 33.46 # 1
SSIM 0.8932 # 2
JPEG Artifact Correction Classic5 (Quality 40 Grayscale) Residual Dense Network + PSNR 34.29 # 1
SSIM 0.9063 # 1
Color Image Denoising Kodak24 sigma10 Residual Dense Network + PSNR 37.33 # 1
Grayscale Image Denoising Kodak24 sigma10 Residual Dense Network + PSNR 35.19 # 1
Color Image Denoising Kodak24 sigma30 Residual Dense Network + PSNR 31.98 # 1
Grayscale Image Denoising Kodak24 sigma30 Residual Dense Network + PSNR 30.02 # 1
Grayscale Image Denoising Kodak24 sigma50 Residual Dense Network + PSNR 27.88 # 1
Color Image Denoising Kodak24 sigma50 Residual Dense Network + PSNR 29.7 # 1
Color Image Denoising Kodak24 sigma70 Residual Dense Network + PSNR 28.24 # 1
Grayscale Image Denoising Kodak24 sigma70 Residual Dense Network + PSNR 26.57 # 1
JPEG Artifact Correction Live1 (Quality 10 Grayscale) Residual Dense Network + PSNR 29.7 # 2
SSIM 0.8252 # 6
JPEG Artifact Correction LIVE1 (Quality 20 Grayscale) Residual Dense Network + PSNR 32.1 # 1
SSIM 0.8886 # 8
JPEG Artifact Correction LIVE1 (Quality 30 Grayscale) Residual Dense Network + PSNR 33.54 # 1
SSIM 0.9156 # 2
JPEG Artifact Correction LIVE1 (Quality 40 Grayscale) Residual Dense Network + PSNR 34.54 # 1
SSIM 0.9304 # 1
Grayscale Image Denoising Urban100 sigma10 Residual Dense Network + PSNR 35.45 # 1
Color Image Denoising Urban100 sigma10 Residual Dense Network + PSNR 36.75 # 1
Grayscale Image Denoising Urban100 sigma30 Residual Dense Network + PSNR 30.08 # 1
Color Image Denoising Urban100 sigma30 Residual Dense Network + PSNR 31.78 # 1
Color Image Denoising Urban100 sigma50 Residual Dense Network + PSNR 29.38 # 1
Grayscale Image Denoising Urban100 sigma50 Residual Dense Network + PSNR 27.47 # 2
Color Image Denoising Urban100 sigma70 Residual Dense Network + PSNR 27.74 # 1
Grayscale Image Denoising Urban100 sigma70 Residual Dense Network + PSNR 25.71 # 1

Methods