Learning to Deblur

By Christian J. Schuler, Michael Hirsch, Stefan Harmeling, Bernhard Schölkopf. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2014

589 citations recorded by the source corpus.

We describe a learning-based approach to blind image deconvolution. It uses a deep layered architecture, parts of which are borrowed from recent work on neural network learning, and parts of which incorporate computations that are specific to image deconvolution. The system is trained end-to-end on a set of artificially generated training examples, enabling competitive performance in blind deconvolution, both with respect to quality and runtime.

Subfields: Computer Vision and Pattern Recognition, Media Technology

View Paper: https://www.semanticscholar.org/paper/93ba9c51439c6632219328419d3619c3df77c135

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

Dirty Pixels: Optimizing Image Classification Architectures for Raw Sensor Data (2017, methodology): “Structured neural networks have been proposed for deblurring [Xu et al. 2014; Schuler et al. 2014; Chakrabarti 2016; Zhang et al. 2016a], denoising [Zhang et al. 2016b], and demosaicking [Gharbi et al. 2016].”
Deep Image Deblurring: A Survey (2022, background): “Early blind deblurring methods focus on removing uniform blur (Cho and Lee 2009; Fergus et al. 2006; Michaeli and Irani 2014; Schuler et al. 2015; Xu and Jia 2010).”
A Neural Approach to Blind Motion Deblurring (2016, methodology): “[16] tackled the general blind motion deblurring problem using a neural architecture designed to mimic the computational steps of traditional iterative deblurring methods.”
AOD-Net: All-in-One Dehazing Network (2017, background): “Examples include image denoising [29], deblurring [20], and super resolution [28].”
A Sparse-View CT Reconstruction Method Based on Combination of DenseNet and Deconvolution (2018, methodology): “Previously, deconvolution, in combination with convolution, has been applied successfully in sematic segmentation problems [47], [49], [50].”
A Dynamic Convolutional Layer for short rangeweather prediction (2015, background): “However, deep learning networks can be used for super resolution [10], blindimage deconvolution [23], noise removal, and other image processing tasks for which the output is an image.”
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].”
Artificial intelligence in the creative industries: a review (2020, unknown): “ral iterations. As described by the image degradation model, the PSF (h) is related to the target image via a convolution operation. CNNs are therefore inherently applicable for solving blur problems [107]. Deblurring techniques based on CNNs [108] and GANs [113] usually employ residual blocks, wher”
Deep Joint Rain Detection and Removal from a Single Image (2016, background): “These applications include denoising [35, 6, 7, 21, 1], completion [38], super-resolution [11, 12, 10, 31], deblurring [32], deconvolution [39] and style transfer [17, 40], etc.”
LLNet: A deep autoencoder approach to natural low-light image enhancement (2015, methodology): “The network was applied for inpainting (Xie et al., 2012) and deblurring (Schuler et al., 2014).”
Joint Rain Detection and Removal from a Single Image with Contextualized Deep Networks (2020, background): “These applications include denoising [23], image completion [24], super-resolution [25], [26], [27], [28], [29], [30], deblur-ring [31], deconvolution [32], style transfer [33], compression artifacts removal [34], [35], etc.”
Convolutional Neural Networks for Inverse Problems in Imaging: A Review (2017, methodology): “The concept of unrolling can also be applied at a coarser scale, as in [13], where the modules of the network mimic the steps of a typical blind deconvolution pipeline: extract features, estimate kernel, estimate image, repeat.”
Scale-Recurrent Network for Deep Image Deblurring (2018, methodology): “Early methods [29, 34, 38] substitute a few modules or steps in traditional frameworks with learned parameters to make use of external data.”
Using Deep Neural Networks for Inverse Problems in Imaging: Beyond Analytical Methods (2018, unknown): “[51] perform end-to-end training of a deep layered architecture of which the first layers, corresponding to the feature extraction step, are learned through training a CNN, but the other two modules of the architecture are fixed and correspond to operations implemented by traditional image deconvolu”
Sparse Gradient Regularized Deep Retinex Network for Robust Low-Light Image Enhancement (2021, background): “These applications include denoising [25], completion [26], inpainting [27], super-resolution [28], deblurring [29], deconvolution [30], and style transfer [31], etc .”
Online Video Deblurring via Dynamic Temporal Blending Network (2017, background): “Previous learning-based single-image blind deconvolution [3, 28, 32] and burst deblurring [35] approaches have considered only ego motion and assumed a uniform blur model.”
Neural Blind Deconvolution Using Deep Priors (2019, methodology): “Given the training set, deep CNNs can either be used to extract features to facilitate blur kernel estimation [1, 38], or be deployed to learn the direct mapping to clean image for motion de-blurring [7,25,43,51].”
Deep Stacked Hierarchical Multi-Patch Network for Image Deblurring (2019, methodology): “Recently, CNNs have been used in non-uniform image deblurring to deal with the complex motion blur in a time-efﬁcient manner [17, 18, 21, 24, 25, 28].”