Learning Data Terms for Non-blind Deblurring

Dong, Jiangxin; Pan, Jinshan; Sun, Deqing; Su, Zhixun; Yang, Ming–Hsuan

doi:10.1007/978-3-030-01252-6_46

Cited by 36 publications

(24 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The estimated inliers are used to define the data term. To obtain more flexible and robust data terms, Dong et al [11] learn discriminative shrinkage functions to indirectly model complex noise distributions, inspired by work on learning regularizers [36].…”

Section: Related Workmentioning

confidence: 99%

“…One family of methods focuses on advancing the data term to better measure the image reconstruction error. Starting from the most commonly used ℓ 2 norm [17], data terms have been carefully designed for specific types of outliers [1,6] or even discriminatively learned [11,28]. A second family of approaches focuses on developing effective regularization terms/image priors to ensure desirable properties of the estimated clear image.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Learning Spatially-Variant MAP Models for Non-blind Image Deblurring

Dong

Roth

Schiele

2021

2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

Self Cite

View full text Add to dashboard Cite

The classical maximum a-posteriori (MAP) framework for non-blind image deblurring requires defining suitable data and regularization terms, whose interplay yields the desired clear image through optimization. The vast majority of prior work focuses on advancing one of these two crucial ingredients, while keeping the other one standard. Considering the indispensable roles and interplay of both data and regularization terms, we propose a simple and effective approach to jointly learn these two terms, embedding deep neural networks within the constraints of the MAP framework, trained in an end-to-end manner. The neural networks not only yield suitable image-adaptive features for both terms, but actually predict per-pixel spatially-variant features instead of the commonly used spatially-uniform ones. The resulting spatially-variant data and regularization terms particularly improve the restoration of fine-scale structures and detail. Quantitative and qualitative results underline the effectiveness of our approach, substantially outperforming the current state of the art.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Learning Spatially-Variant MAP Models for Non-blind Image Deblurring

Dong

Roth

Schiele

2021

2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

Self Cite

View full text Add to dashboard Cite

show abstract

“…The maximum dilation rate we set for each FRC (Fig.2) is 4, so the order of dilation rates in each FRC block is (1, 2, 3, 4, 3, 2, 1). Note that when the dilation rate equals 1, the corresponding size of the convolution kernel is (3,3). The output of FRC will enter into an SA or BA module.…”

Section: B Details Of Architecture and Training Processmentioning

confidence: 99%

“…In Table II, there're 5 models with different proposed modules and for the models (model1, model2 and model3) without FRC modules, we have designed an encoder-decoder module for each of them. Note that, the encoder-decoder module has the same number of parameters as FRC module because they have the same number of convolutional layers whose kernels are in the shape of (3,3). From model1 and model2 we can find that the SA module can improve the performance of the network without producing extra parameters.…”

Section: B Details Of Architecture and Training Processmentioning

confidence: 99%

Multi-Frame Blind Restoration for Image of Space Target With FRC and Branch-Attention

2020

View full text Add to dashboard Cite

The random noise and anisotropic motion of atmospheric turbulence can cause different degradation patterns, which make images of space targets observed from ground-based stations severely disturbed. In recent years, benefit from the development of convolutional neural networks (CNNs), a large number of effective end-to-end methods were proposed to restore images. However, a single-frame method whose input is just a single image can hardly achieve a further improvement for the restoration image due to the diversified degradation patterns of space-target images. In this paper, we proposed a multi-branch network with a multi-frame input to restore space-target images. The multi-frame input contains spacetarget images which own different degradation patterns at different moments. In this way, we can fully use the complementary information between input frames. And in this network, two effective technologies are introduced: one is the full resolution convolution module which extracts features by using convolutional layers with different dilation rates to keep feature information complete; the other is the branch-attention module which is used to pass effective information between different branches of the network. Furthermore, we demonstrated the effectiveness of our method by comparing it with those state-of-the-art methods. INDEX TERMS Multi-frame image restoration, multi-branch network, self-attention, branch-attention, full resolution convolution, images of space targets.

show abstract

“…Recently, a wide variety of efficient methods based on MAP framework have been proposed. Many of them focused on designing different data terms [4,24] and various kinds of image priors [3,5,7,[25][26][27][28]. In addition, patch-based methods [13,29,30] have been developed to sidestep classical regularizers and had shown impressive performance.…”

Section: Related Workmentioning

confidence: 99%

Blind Deconvolution with Scale Ambiguity

Fan

Wang

et al. 2020

Applied Sciences

Self Cite

View full text Add to dashboard Cite

Recent years have witnessed significant advances in single image deblurring due to the increasing popularity of electronic imaging equipment. Most existing blind image deblurring algorithms focus on designing distinctive image priors for blur kernel estimation, which usually play regularization roles in deconvolution formulation. However, little research effort has been devoted to the relative scale ambiguity between the latent image and the blur kernel. The well-known L 1 normalization constraint, i.e., fixing the sum of all the kernel weights to be one, is commonly selected to remove this ambiguity. In contrast to this arbitrary choice, we in this paper introduce the L p -norm normalization constraint on the blur kernel associated with a hyper-Laplacian prior. We show that the employed hyper-Laplacian regularizer can be transformed into a joint regularized prior based on a scale factor. We quantitatively show that the proper choice of p makes the joint prior sufficient to favor the sharp solutions over the trivial solutions (the blurred input and the delta kernel). This facilitates the kernel estimation within the conventional maximum a posterior (MAP) framework. We carry out numerical experiments on several synthesized datasets and find that the proposed method with p = 2 generates the highest average kernel similarity, the highest average PSNR and the lowest average error ratio. Based on these numerical results, we set p = 2 in our experiments. The evaluation on some real blurred images demonstrate that the results by the proposed methods are visually better than the state-of-the-art deblurring methods.

show abstract

Learning Data Terms for Non-blind Deblurring

Cited by 36 publications

References 25 publications

Learning Spatially-Variant MAP Models for Non-blind Image Deblurring

Learning Spatially-Variant MAP Models for Non-blind Image Deblurring

Multi-Frame Blind Restoration for Image of Space Target With FRC and Branch-Attention

Blind Deconvolution with Scale Ambiguity

Contact Info

Product

Resources

About