Unnatural L0 Sparse Representation for Natural Image Deblurring

Liu, Xü; Zheng, Shicheng; Jia, Jiaya

doi:10.1109/cvpr.2013.147

Cited by 861 publications

(968 citation statements)

References 29 publications

Supporting

Mentioning

956

Contrasting

Unclassified

Order By: Relevance

“…The latter methods aim at approximating the L 0 "pseudo-norm" of the gradients, as proposed also in (Xu et al 2013). In this paper we also encourage sparsity in the gradients.…”

Section: Prior Workmentioning

confidence: 99%

“…In the past decade, several high-performing blind deconvolution schemes using Bayesian principles have been proposed (Babacan et al 2012;Cho and Lee 2009;Fergus et al 2006;Krishnan et al 2011Krishnan et al , 2013Levin et al 2011a;Shan et al 2008;Wipf and Zhang 2014;Xu and Jia 2010;Xu et al 2013). The first step in the Bayesian framework is to devise a statistical distribution for both the gradients of the sharp image and the measurement noise or the model error.…”

Section: Prior Workmentioning

confidence: 99%

“…Solving blind deconvolution is a challenging task because it is a non-convex and ill-posed problem that requires the estimation of both u and k. For this reason it is often tackled with the use of regularization by means of prior knowledge on the distribution of sharp images (Babacan et al 2012;Cho and Lee 2009;Fergus et al 2006;Krishnan et al 2011Krishnan et al , 2013Levin et al 2011a;Shan et al 2008;Wipf and Zhang 2014;Xu and Jia 2010;Xu et al 2013). Several methods use also a prior on the blur kernel (see for instance the recent works (Keuper et al 2013;Kenig et al 2010)).…”

mentioning

confidence: 99%

See 2 more Smart Citations

A Logarithmic Image Prior for Blind Deconvolution

Perrone

Favaro

2015

Int J Comput Vis

View full text Add to dashboard Cite

Blind Deconvolution consists in the estimation of a sharp image and a blur kernel from an observed blurry image. Because the blur model admits several solutions it is necessary to devise an image prior that favors the true blur kernel and sharp image. Many successful image priors enforce the sparsity of the sharp image gradients. Ideally the L 0 "norm" is the best choice for promoting sparsity, but because it is computationally intractable, some methods have used a logarithmic approximation. In this work we also study a logarithmic image prior. We show empirically how well the prior suits the blind deconvolution problem. Our analysis confirms experimentally the hypothesis that a prior should not necessarily model natural image statistics to correctly estimate the blur kernel. Furthermore, we show that a simple Maximum a Posteriori (MAP) formulation is enough to achieve state of the art results. To minimize such formulation we devise two iterative minimization algorithms that cope with the non-convexity of the logarithmic prior: one obtained via the primaldual approach and one via majorization-minimization.

show abstract

“…The latter methods aim at approximating the L 0 "pseudo-norm" of the gradients, as proposed also in (Xu et al 2013). In this paper we also encourage sparsity in the gradients.…”

Section: Prior Workmentioning

confidence: 99%

Section: Prior Workmentioning

confidence: 99%

mentioning

confidence: 99%

See 1 more Smart Citation

A Logarithmic Image Prior for Blind Deconvolution

Perrone

Favaro

2015

Int J Comput Vis

View full text Add to dashboard Cite

show abstract

“…Krishnan et al [9] used l 1 /l 2 norm as the sparsity measure which acts as a scale invariant regularizer. Xu et al [10] introduced L 0 sparsity as regularization term, which relies on the intermediate representation of image for its success. Pan et.…”

Section: Iirelated Workmentioning

confidence: 99%

“…The prior MAP based approaches that are successful can be roughly classified into two types i.e. those with explicit edge prediction steps like using a shock filter [2,3,4,5,6,7,8] and those which include implicit regularization process [9,10]. The common factor is the intermediate image consisting of only step like structures called the unnatural representation of the image, which is the key factor in the success of MAP based methods.…”

Section: Iintroductionmentioning

confidence: 99%

Handling Noise and Outliers in Single Image Deblurring using L0 Sparsity

Ashok¹

2015

IJAREEIE

View full text Add to dashboard Cite

Camera shake during exposure leads to image blur and poses an important problem in digital photography. Blind deconvolution recovers the sharp original image from a blurred image. MAP has been the most widely used deconvolution field but naive MAP methods mostly tends to favour no-blur solution. An intermediate representation of the image called unnatural representation has been found to the main reason for success of existing MAP based methods. This paper presents a new blind deconvolution algorithm based on L 0 sparsity for single image motion deblurring, which is robust to the presence of noise. Directional filtering is used for noise handling process owing to the fact that it does not interfere with blur of the image affecting the kernel estimation process adversely. Also, a non-blind deconvolution step with explicit outlier handling is incorporated in final image restoration step which ensures that the image is devoid of ringing artifacts and handles other non-linearities. This presents an efficient optimization problem which requires only a few iterations solve and provides very good visual quality images with no ringing artifacts despite the presence of noise. The results are comparable to state-of-the-art methods dealing with images affected with only blur while being robust to noise unlike state-of-the-art methods. KEYWORDS:Camera shake, motion blur,L 0 sparsity, directional filtering, outlier handling, ringing artifacts. I.INTRODUCTIONMotion blur is a common artifact in digital photography which produces blurry images with inevitable information loss. Taking handheld photographs in low light condition is a challenging process. Higher exposure times are needed due to the lesser availability of light in this case and photos end up blurry because the photographer moves during exposure time. Usage of the tripod and mechanical stabilization of lens or sensors is a likely solution, but transportation issues and other factors limit the usefulness and longer exposure time remains a challenging problem. Increasing light sensitivity of the camera using a higher ISO setting can reduce the exposure time but a trade off with noise exists in this case. Reduction in the exposure time comes at the cost of increase in the noise levels but the noise levels still remains considerably high for handheld photography. As a result, the photos end up being blurry and noisy. The process of recovering the original image from a blurred image has been a fundamental research problem of major interest in the field of digital image processing. The recent years have witnessed significant advances in the development of single image deblurring techniques. Image deblurring techniques can be primarily divided into two types-blind image deblurring and non-blind image deblurring. In non-blind deconvolution process, the motion blur kernel is assumed to be known and we have to just estimate the latent image. Blind deconvolution is a much more ill posed problem ,as both the blur kernel that has caused the blurring and the latent imag...

show abstract

Materials Research at Shanghai Jiao Tong University

Chen

Feng

2015

Advanced Materials

View full text Add to dashboard Cite

Transformer architectures have exhibited remarkable performance in image super-resolution (SR). Since the quadratic computational complexity of the self-attention (SA) in Transformer, existing methods tend to adopt SA in a local region to reduce overheads. However, the local design restricts the global context exploitation, which is critical for accurate image reconstruction. In this work, we propose the Recursive Generalization Transformer (RGT) for image SR, which can capture global spatial information and is suitable for high-resolution images. Specifically, we propose the recursive-generalization self-attention (RG-SA). It recursively aggregates input features into representative feature maps, and then utilizes cross-attention to extract global information. Meanwhile, the channel dimensions of attention matrices (query, key, and value) are further scaled for a better trade-off between computational overheads and performance. Furthermore, we combine the RG-SA with local self-attention to enhance the exploitation of the global context, and propose the hybrid adaptive integration (HAI) for module integration. The HAI allows the direct and effective fusion between features at different levels (local or global). Extensive experiments demonstrate that our RGT outperforms recent state-of-the-art methods.

show abstract

Unnatural L0 Sparse Representation for Natural Image Deblurring

Cited by 861 publications

References 29 publications

A Logarithmic Image Prior for Blind Deconvolution

A Logarithmic Image Prior for Blind Deconvolution

Handling Noise and Outliers in Single Image Deblurring using L0 Sparsity

Materials Research at Shanghai Jiao Tong University

Contact Info

Product

Resources

About