Accurate Blur Models vs. Image Priors in Single Image Super-resolution

By Netalee Efrat, Daniel Glasner, Alexander Apartsin, Boaz Nadler, Anat Levin. 2013

179 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Radiation

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 Radiation

Citations in the corpus

Designing a Practical Degradation Model for Deep Blind Image Super-Resolution (2021, background): “The reason is that blur kernels play a vital role for the success of SISR methods [12] and a bicubic kernel is too simple.”
Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network (2016, neutral): “Recent popular SISR methods can be classified into edge-based [35], image statisticsbased [9, 18, 46, 12] and patch-based [2, 43, 52, 13, 54, 40, 5] methods.”
Deep Unfolding Network for Image Super-Resolution (2020, background): “…various LR images for an HR image, with different blur kernels, scale factors and noise, the study of learning a single deep model to invert all such LR images to HR image is still lacking. simplistic degradation assumption of existing SISR meth-ods and the complex degradations of real images [16].”
Flow-based Kernel Prior with Application to Blind Super-Resolution (2021, methodology): “If the estimated kernel deviates from the ground-truth, the HR image reconstructed by the non-blind SR methods would deteriorate seriously [11, 17, 50].”
Zero-Shot Super-Resolution Using Deep Internal Learning (2017, result): “This supports the analysis and observations of [18], that (i) an accurate downscaling model is more important than sophisticated image priors, and (ii) using the wrong donwscaling kernel leads to oversmoothed SR results.”
Conditioned Regression Models for Non-blind Single Image Super-Resolution (2015, unknown): “In the field of image deconvolution, this is actually the standard setup and many different approaches exist to handle varying blur kernels during inference.”
Blind image super-resolution with spatially variant degradations (2019, background): “The relative importance of image prior and reconstruction constraint was investigated by Efrat et al. [2013] who showed the importance of blur kernel estimation in the SR problem.”
Blind Super-Resolution With Iterative Kernel Correction (2019, background): “As has been revealed in [9, 36], learning-based methods will suffer severe performance drop when the pre-deﬁned blur kernel is different from the real one.”
Blind Super-Resolution Kernel Estimation using an Internal-GAN (2019, background): “It was shown in [26] that the effect of using an incorrect SR-kernel is of greater”
Seven Ways to Improve Example-Based Single Image Super Resolution (2015, result): “Note that generally the explicit form for the degradation operators is unknown and their estimation from prior train examples is imprecise [11, 26], therefore our reported results with (B) refinement are an upper bound for a practical implementation and difficult to reach.”
Real-World Single Image Super-Resolution: A Brief Review (2021, methodology): “As highlighted in [141], most of the existing degradation modeling-based methods focus on the recovery of the blur kernel.”
Plug-and-Play Image Restoration With Deep Denoiser Prior (2020, methodology): “where ↓ s denotes bicubic downsamling with downscaling factor s, it has been revealed that these methods would deteriorate seriously if the real degradation model deviates from the assumed one [76], [77].”
Deep Learning for Single Image Super-Resolution: A Brief Review (2018, background): “, [57], have shown that when an inaccurate estimation is input, there are side effects on the final performance.”