Natural image denoising: Optimality and inherent bounds

By Anat Levin, Boaz Nadler. 2011

264 citations recorded by the source corpus.

Subfields: Computational Mechanics, 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 Computational Mechanics

In Computer Vision and Pattern Recognition

In Media Technology

Citations in the corpus

The Perception-Distortion Tradeoff (2017, background): “Several past works attempted to answer the question: What is the minimal attainable distortion Dmin in various restoration tasks? [49], [50], [51], [52], [53].”
Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising (2016, methodology): “Moreover, high noise level usually requires larger effective patch size to capture more context information for restoration [34].”
Universal Denoising Networks : A Novel CNN Architecture for Image Denoising (2017, background): “However, during the past decade there was little progress in improving the state-of-the art denoising performance and it has been suggested that denoising algorithms have reached optimality and cannot be further improved [28].”
Nonlocal Transform-Domain Filter for Volumetric Data Denoising and Reconstruction (2013, background): “Recent works discuss the near-optimality of this approach and offer further insights about the rationale of the algorithm [3], [4].”
Trainable Nonlinear Reaction Diffusion: A Flexible Framework for Fast and Effective Image Restoration (2015, methodology): “This is a notable improvement as few methods can surpass BM3D more than 0.3dB in average [35].”
Image denoising with multi-layer perceptrons, part 2: training trade-offs and analysis of their mechanisms (2012, background): “This is in agreement with Levin and Nadler (2011): Using a larger support size makes the denoising problem less ambiguous.”
A Nonlocal Bayesian Image Denoising Algorithm (2013, methodology): “Two types of performance bounds proposed in [28] and [8] address the same class of patchbased algorithms and propose denoising methods approaching these bounds (section 3).”
Weighted Nuclear Norm Minimization and Its Applications to Low Level Vision (2016, methodology): “Such an improvement is notable since few methods can surpass BM3D more than 0.3dB on average (Levin et al, 2012; Levin and Nadler, 2011).”
Deep Mean-Shift Priors for Image Restoration (2017, methodology): “Note that this is the same loss as in non-parametric least squares estimators [23, 26, 20].”
Singular vector and singular subspace distribution for the matrix denoising model (2018, background): “For instance, in the study of image denoising, S is treated as the true image [50] and in the problem of classification, S contains the the underlying true mean vectors of samples [17].”
Image denoising: Can plain neural networks compete with BM3D? (2012, result): “This is in agreement with previous findings [12].”
PointCleanNet: Learning to Denoise and Remove Outliers from Dense Point Clouds (2019, background): “In signal processing literature, it is widely believed that image de-noising has reached close to optimal performance [CM10,LN11].”
Real Image Denoising With Feature Attention (2019, unknown): “[37] and Chatterjee et al .”
FFDNet: Toward a Fast and Flexible Solution for CNN-Based Image Denoising (2017, background): “Further increase of receptive field actually benefits little in improving denoising performance [40].”
The Little Engine That Could: Regularization by Denoising (RED) (2016, background): “Indeed, these documented results have led researchers to the educated guess that these methods are getting very close to the optimally possible denoising performance [21, 22, 23].”
Patch Complexity, Finite Pixel Correlations and Optimal Denoising (2012, methodology): “1 Empirical study To understand the dependence of PSNR on window size, we prese nt an mpirical study with M = 10(4) clean and noisy pairs {(xj , yj)}j=1 andN = 10(8) samples taken from the LabelMe dataset, as in [14].”
An Analysis and Implementation of the BM3D Image Denoising Method (2012, background): “It is furthermore interesting to notice that recent papers [3, 14] which try to evaluate the inherent bounds of patch-based denoising methods claim that BM3D is really close to those optimality bounds.”
On learning optimized reaction diffusion processes for effective image restoration (2015, methodology): “This is a notable im-q JPEG decoder TGV [5] Dic-SR[6] SADCT[15] RTF[ provement as few methods can surpass BM3D more than 0.3dB in average [26].”