Fast Image Deconvolution using Hyper-Laplacian Priors

By Dilip Krishnan, Rob Fergus. Neural Information Processing Systems, 2009

1,474 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/9009461b81bed5dbf69432ff679f9fba1ef7adc5

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Learning Proximal Operators: Using Denoising Networks for Regularizing Inverse Imaging Problems (2017, unknown): “51 LUT [23] 24.”
From learning models of natural image patches to whole image restoration (2011, methodology): “alternative optimization method called “Half Quadratic Splitting” which has been proposed recently in several relevant contexts [11, 7].”
Image Restoration Using Convolutional Auto-encoders with Symmetric Skip Connections (2016, methodology): “kernel tpye Krishnan [57] Levin [58] Cho [59] Schuler [60] Xu [56] RED30”
Hierarchical Convolutional Deep Learning in Computer Vision (2013, methodology): “To optimize this, we experimented with closed form optimization techniques in [64], but settled on gradient descent for simplicity.”
Deep Convolutional Neural Network for Inverse Problems in Imaging (2016, background): “2713099 problems in imaging including denoising [1]–[3], deconvolution [4], [5], and interpolation [6].”
Learning a convolutional neural network for non-uniform motion blur removal (2015, background): “Image deblurring [2, 6, 9, 11, 14, 19, 23, 30] aims at recovering sharp image from a blurry image due to camera shake, object motion or out-of-focus.”
Proximal Deep Structured Models (2016, background): “While the original proximal operator was designed for the purpose of obtaining the global optimum in convex optimization, recent work has shown that proximal methods work well for non-convex optimization as long as the proximal operator exists [20, 33, 6].”
Blind deconvolution in scientific imaging & computational photography (2011, methodology): “For the final non-blind deconvolution we set the exponent α to 2/3 as it allows the use of an analytical solution for the w-subproblem (see (Krishnan and Fergus, 2009) for details).”
Fast removal of non-uniform camera shake (2011, methodology): “Choosing α = 2/3 allows an analytical formula for the update in v, see [9] for details.”
3D Neural Field Generation Using Triplane Diffusion (2022, background): “[4] found that gradients of natural images are closely modelled by a hyper-Laplacian with 0.”
Non-uniform Deblurring for Shaken Images (2010, methodology): “For the results of single-image deblurring in Section 5, we have adapted the deconvolution algorithm of Krishnan and Fergus (2009), which performs MAP estimation of the sharp image using a hyper-Laplacian prior on the image gradients.”
Efficient marginal likelihood optimization in blind deconvolution (2011, background): “One can also consider the fast solver of [9], but we found that for this application, conjugate gradient converges faster.”
Understanding deep image representations by inverting them (2014, background): “Hyper-Laplacian priors with β < 1 are often used as a better match of the gradient statistics of natural images [14], but they only exacerbate this issue.”
Deep Mean-Shift Priors for Image Restoration (2017, unknown): “ovide a summary in Table 1. For brevity, we omit the role of downsampling (required for super-resolution) and masking. 5 Levin [19] Berkeley [2] Method ˙ n: 2.55 5.10 7.65 10.2 2.55 5.10 7.65 10.2 FD [18] 30.03 28.40 27.32 26.52 24.44 23.24 22.64 22.07 EPLL [39] 32.03 29.79 28.31 27.20 25.38 23.53 2”
Visualizing Deep Convolutional Neural Networks Using Natural Pre-images (2015, methodology): “Perhaps the first to apply activation maximization to recent CNNs such as AlexNet (Krizhevsky et al. 2012) was the work of Simonyan et al. (2014), where this technique was used tomaximize the response of neural activations in the last layer of a deepCNN.”
Deblurring Text Images via L0-Regularized Intensity and Gradient Prior (2014, methodology): “We note that non-blind deblurring methods with Laplacian priors [9] have been shown to preserve fine details.”