Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising
By Kai Zhang, Wangmeng Zuo, Yunjin Chen, Deyu Meng, Lei Zhang. 2016
8,505 citations recorded by the source corpus.
The discriminative model learning for image denoising has been recently attracting considerable attentions due to its favorable denoising performance. In this paper, we take one step forward by investigating the construction of feed-forward denoising convolutional neural networks (DnCNNs) to embrace the progress in very deep architecture, learning algorithm, and regularization method into image denoising. Specifically, residual learning and batch normalization are utilized to speed up the training process as well as boost the denoising performance. Different from the existing discriminative denoising models which usually train a specific model for additive white Gaussian noise at a certain noise level, our DnCNN model is able to handle Gaussian denoising with unknown noise level (i.e., blind Gaussian denoising). With the residual learning strategy, DnCNN implicitly removes the latent clean image in the hidden layers. This property motivates us to train a single DnCNN model to tackle with several general image denoising tasks, such as Gaussian denoising, single image super-resolution, and JPEG image deblocking. Our extensive experiments demonstrate that our DnCNN model can not only exhibit high effectiveness in several general image denoising tasks, but also be efficiently implemented by benefiting from GPU computing.
Subfields: Computer Vision and Pattern Recognition, Media Technology
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Computer Vision and Pattern Recognition
- Luc Van Gool
- Ming-Hsuan Yang
- Dacheng Tao
- Lei Zhang
- Chen Change Loy
- Xiaogang Wang
- Trevor Darrell
- Chunhua Shen
In Media Technology
Citations in the corpus
- Designing a Practical Degradation Model for Deep Blind Image Super-Resolution (2021, methodology): “For example, DnCNN [50] is able to handle SISR with different scale factors, JPEG compression deblocking with different quality factors and denoising for a wide range of noise levels, while still having a performance comparable to VDSR [21] for SISR.”
- Memory-Efficient Hierarchical Neural Architecture Search for Image Denoising (2019, methodology): “Recent network models such as DnCNN [41] and IrCNN [42] predict the residue presented in the image instead of the denoised image, showing promising performance.”
- Palette: Image-to-Image Diffusion Models (2021, positive): “[2018b] train JPEG restoration models on multiple quality factors and [Zhang et al., 2017] train denoising models with a large range of noise level degradation.”
- Enhancing Evolutionary Couplings with Deep Convolutional Neural Networks (2017, background): “In the image recognition field, CNNs have proven very effective at taking a noisy image and returning the clean image (Zhang et al., 2017).”
- Solving Linear Inverse Problems Using the Prior Implicit in a Denoiser (2020, methodology): “For a denoiser, we used BF-CNN [38], a bias-free variant of DnCNN [35].”
- Pre-Trained Image Processing Transformer (2020, background): “DnCNN [75] MemNet [57] IRCNN [76] FFDNet [77] IPT (ours)”
- SinGAN: Learning a Generative Model From a Single Natural Image (2019, background): “The role of the convonlutional layers is to generate the missing details in (x̃n+1) ↑r (residual learning [22, 57]).”
- Learning Enriched Features for Real Image Restoration and Enhancement (2020, methodology): “al-resolution reduction, on the downside, the ne spatial details are lost, making it extremely hard to recover them in the later stages. On the other side, the high-resolution (single-scale) networks [27,120,127,50] do not employ any downsampling operation, and thereby produce images with spatially ”
- Learning Proximal Operators: Using Denoising Networks for Regularizing Inverse Imaging Problems (2017, methodology): “We use the training pipeline identical to [47] with the Adam optimization algorithm [20] and train our network for removing Gaussian noise of a fixed standard deviation σ.”
- Noise2Self: Blind Denoising by Self-Supervision (2019, methodology): “As shown in Table 2, we achieve superior performance to the classic unsupervised denoisers NLM and BM3D, and comparable performance to the same architecture trained with clean targets (Noise2Truth) and with independently noisy targets (Noise2Noise), and to a purely convolutional architecture with cl”
- CycleISP: Real Image Restoration via Improved Data Synthesis (2020, methodology): “Several recent methods use deep CNNs [4, 7, 25, 28, 45, 64, 65, 2] and demonstrate promising denoising performance.”
- GenImage: A Million-Scale Benchmark for Detecting AI-Generated Image (2023, methodology): “The noise residuals R can be computed by the formula R = X − F ( X ) , where X is the input image, and F ( · ) is the denoising filter DCNN [37].”
- Image denoising using deep CNN with batch renormalization (2020, methodology): “Although the above methods have shown excellent performance in image denoising, most of these methods are faced with two major challenges [61]: (1) manual tuning of the penalty parameters, and (2) complex optimized algorithms.”
- Using labeled data to evaluate change detectors in a multivariate streaming environment (2009, neutral): “We denote the number of input, output channels as C in , C out , and the number of output channels of convolutional layers and maxpooling layers as C con v , C maxp .”
- Convolutional Neural Networks for Inverse Problems in Imaging: A Review (2017, methodology): “The aim of this constrained scope is to allow us to draw meaningful generalizations from the surveyed works. denoising deconvolution super-resolution MRI CT [6]–[11] [10], [12]–[14] [9], [15]–[20] [21]–[23] [24]–[27] The manuscript is organized as follows.”
- Second-Order Adversarial Attack and Certifiable Robustness (2018, background): “However, we note that there have been notable efforts at developing robust DNNs to natural perturbations Xie et al. (2012); Zhang et al. (2017); yet, these methods failed to defend models from adversarial attacks, as they are not particularly designed for that task.”