Natural Image Denoising with Convolutional Networks

By Viren Jain, Sebastian Seung. 2008

982 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition

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

Citations in the corpus

Instance-Sensitive Fully Convolutional Networks (2016, background): “These fully convolutional models are naturally applicable for image restoration problems, such as denoising [16], super-resolution [17], and others [18], where each output pixel is a real-number regressor of intensity values.”
Representation Learning: A Review and New Perspectives (2012, methodology): “An early approach (Jain and Seung, 2008) trained a standard but deep convolutional MLP on the task of denoising images, i.e. as a deep, convolutional, denoising auto-encoder.”
Deep learning in neural networks: An overview (2014, methodology): “To detect human actions in surveillance videos, a 3-dimensional CNN (e.g., Jain & Seung, 2009; Prokhorov, 2010), combined with SVMs, was part of a larger system (Yang et al., 2009) using a bag of features approach (Nowak, Jurie, & Triggs, 2006) to extract regions of interest.”
What regularized auto-encoders learn from the data-generating distribution (2012, background): “…some form of reconstruction error , such as auto-encoder and sparse coding variants (Olshausen and Field, 1997; Bengio et al., 2007; Ranzato et al., 2007; Jain and Seung, 2008; Ranzato et al., 2008; Vincent et al., 2008; Kavukcuoglu et al., 2009; Rifai et al., 2011b,a; Gregor et al., 2011).”
Image Super-Resolution Using Deep Convolutional Networks (2014, background): “We empirically find that a smaller learning rate in the last layer is important for the network to converge (similar to the denoising case [21]).”
N4-Fields: Neural Network Nearest Neighbor Fields for Image Transforms (2014, background): “Another series of works [14, 15] investigate the use of convolutional neural networks for image denoising.”
Solving Linear Inverse Problems Using the Prior Implicit in a Denoiser (2020, background): “One expresses the estimation function (as opposed to the prior) in parametric form, and sets the parameters by minimizing the denoising MSE over a large training set of example signals and their noise-corrupted counterparts [23–26].”
Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising (2016, background): “It has been pointed out that the receptive ﬁeld size of denoising neural networks correlates with the effective patch size of denoising methods [23], [24].”
Global-Local Face Upsampling Network (2016, methodology): “These methods have also been replacing highly optimized handdesigned algorithms in low-level vision tasks such as image denoising [20,21,22], image enhancement [23,24], and SR [25,26,27].”
FFDNet: Toward a Fast and Flexible Solution for CNN-Based Image Denoising (2017, methodology): “popular approach is plain discriminative learning, such as the MLP [18] and convolutional neural network (CNN) based methods [19], [20], among which the DnCNN [20] method has achieved very competitive denoising performance.”
On learning optimized reaction diffusion processes for effective image restoration (2015, methodology): “The proposed diffusion model also bears an interesting link to the convolutional networks (CNs) employed for image restoration problems [24].”
Trainable Nonlinear Reaction Diffusion: A Flexible Framework for Fast and Effective Image Restoration (2015, background): “Unfortunately, we are not able to present comparisons with [32], [39], as their codes are not available.”
KIKI‐net: cross‐domain convolutional neural networks for reconstructing undersampled magnetic resonance images (2018, methodology): “The network architecture is based on deep CNNs that have been proven to learn pixel-to-pixel regression in the area of computer vision.(25-32) The CD-CNNs consist of 3 components: a deep CNN for k-space completion (KCNN), a deep CNN for image restoration (ICNN), and an interleaved data consistency (”
Convolutional Neural Network With Data Augmentation for SAR Target Recognition (2016, result): “Meanwhile, CNN does not assume the distribution of noise in the input images, so it can be adaptable to more general and complex noise models, which is consistent with the empirical results in [16].”
Regularized Auto-Encoders Estimate Local Statistics (2012, background): “…some form of reconstruction error, such as auto-encoder and sparse coding variants (Olshausen and Field, 1997; Bengio et al., 2007; Ranzato et al., 2007; Jain and Seung, 2008; Ranzato et al., 2008; Vincent et al., 2008; Kavukcuoglu et al., 2009; Rifai et al., 2011a,b; Gregor et al., 2011).”
Restoring an Image Taken through a Window Covered with Dirt or Rain (2013, background): “Jain and Seung [10], and Zhang and Salari [21].”
Loss Functions for Neural Networks for Image Processing (2015, methodology): “The loss layer, despite being the effective driver of the network’s learning, has at-tracted little attention within the image processing research community: the choice of the cost function generally defaults to the squared 2 norm of the error [8,3,9,10].”
Denoising genome-wide histone ChIP-seq with convolutional neural networks (2016, methodology): “Achieving optimal experimental parameters and comparable data quality across experiments is often difficult, costly, or even impossible, resulting in low sensitivity and specificity of measurements especially in low input samples such as rare populations of primary cells and tissues (Brind’Amour et…”
Dirty Pixels: Optimizing Image Classification Architectures for Raw Sensor Data (2017, methodology): “Conventional fully-connected or convolutional neural networks have also been successfully applied to low-level image processing tasks (see, e.g., [Jain and Seung 2009; Xie et al. 2012; Burger et al. 2012; Dong et al. 2014; Kim et al. 2016]).”