Understanding and evaluating blind deconvolution algorithms

By Anat Levin, Yair Weiss, Frédo Durand, William T. Freeman. 2009 IEEE Conference on Computer Vision and Pattern Recognition, 2009

1,447 citations recorded by the source corpus.

Blind deconvolution is the recovery of a sharp version of a blurred image when the blur kernel is unknown. Recent algorithms have afforded dramatic progress, yet many aspects of the problem remain challenging and hard to understand. The goal of this paper is to analyze and evaluate recent blind deconvolution algorithms both theoretically and experimentally. We explain the previously reported failure of the naive MAP approach by demonstrating that it mostly favors no-blur explanations. On the other hand we show that since the kernel size is often smaller than the image size a MAP estimation of the kernel alone can be well constrained and accurately recover the true blur. The plethora of recent deconvolution techniques makes an experimental evaluation on ground-truth data important. We have collected blur data with ground truth and compared recent algorithms under equal settings. Additionally, our data demonstrates that the shift-invariant blur assumption made by most algorithms is often violated.

Subfields: Computer Vision and Pattern Recognition, Media Technology

View Paper: https://www.semanticscholar.org/paper/262aead5e5db0842807597cc0a81c6e5c265a9a8

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

In Media Technology

Citations in the corpus

Ieee Transactions on Pattern Analysis and Machine Intelligence 1 on Bayesian Adaptive Video Super Resolution (methodology): “Although similar Bayesian MAP approach performed poorly for general debluring problems [18], the spatial smoothness prior on the kernel prevents kernel estimation from converging to the delta function, as shown by [14].”
Deep Unfolding Network for Image Super-Resolution (2020, neutral): “For this reason, as shown in Table 1, we only consider 12 representative and diverse blur kernels, including 4 isotropic Gaussian kernels with different widths ( i.e ., 0.7, 1.2, 1.6 and 2.0), 4 anisotropic Gaussian kernels from [67], and 4 motion blur kernels from [5, 33].”
Training Convolutional Networks with Noisy Labels (2014, background): “When the diffusion kernel is unknown, it is necessary to regularize the ill-posed inverse problem by pushing the estimates away from the trivial identity kernel (Levin et al., 2009 ).”
Hand-Held Video Deblurring Via Efficient Fourier Aggregation (2015, background): “Due to its small spatial support, the blurring kernel estimation is an easier problem to solve than simultaneously estimating both the kernel and the sharp image [21], [22].”
Nonparametric Blind Super-resolution (2013, methodology): “In the context of blind deblurring, this MAPh,k approach has been shown to lead to inaccurate estimates of k [13].”
Image Deblurring via Extreme Channels Prior (2017, methodology): “In addition, we also show the performances of the proposed algorithm against the DCP-based [17] and BCPbased methods on this benchmark dataset [13] in terms of the cumulative error ratio.”
Blind Face Restoration via Deep Multi-scale Component Dictionaries (2020, background): “1 : 5} and 32 motion blur kernels from [1, 25].”
Richardson-Lucy Deblurring for Scenes under a Projective Motion Path (2011, neutral): “However, as recently discussed by Levin et al. [17] this global PSF assumption is typically invalid.”
Fast motion deblurring (2009, methodology): “Quantitative comparison of our method with previous deblurring methods using the data set of [Levin et al. 2009] can be found in the supplementary material.”
Examining the Impact of Blur on Recognition by Convolutional Networks (2016, background): “Graphics and vision researchers have made significant progress on this problem across the last decade [7, 13, 15, 28]—including with recent neural network-based methods [1, 21, 22, 26].”
Learning to Deblur (2014, methodology): “20 shows an example failure case from the benchmark dataset of [26].”
Blind deconvolution in scientific imaging & computational photography (2011, background): “In (Levin et al., 2009) it is already demonstrated that the spatially uniform blur model of most algorithms is an unrealistic assumption, a finding that we found confirmed in our own experiments.”
Deep Convolutional Neural Network for Image Deconvolution (2014, methodology): “The classical Wiener deconvolution is equivalent to using Tikhonov regularizer [2].”
Deep Image Deblurring: A Survey (2022, methodology): “Applying the 8 blur kernels from [64], this results in 640 blurred images.”
Dynamic Scene Deblurring Using Spatially Variant Recurrent Neural Networks (2018, background): “(c) is obtained by blurring (a) with the motion kernel from [16] as shown in (b).”
Deep Plug-And-Play Super-Resolution for Arbitrary Blur Kernels (2019, background): “Motion 32 8 blur kernels from [35] and their augmented 8 kernels by random rotation and flip; and 16 realistic-looking motion blur kernels generated by the released code of [7].”