Fast non-blind deconvolution via regularized residual networks with long/short skip-connections

Son, Hyeongseok; Lee, Seungyong

doi:10.1109/iccphot.2017.7951480

Cited by 40 publications

(34 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Noise level EPLL [52] MLP [38] CSF [36] LDT [11] FCN [49] IRCNN [50] FDN [18] FNBD [39] RGDN [12] (c) MLP [38] (d) CSF [36] (e) LDT [11] (f) FCN [49] (g) IRCNN [50] (h) FDN [18] (i) RGDN [12] (j) SVMAP (ours)…”

Section: Datasetmentioning

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)

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: Datasetmentioning

confidence: 99%

“…The learning rate is initialized as 5 × 10 −5 and halved every 200 epochs. We empirically use M = 3 and N = 5 pixel-dependent filters for the data term and (a) Blurry input (b) EPLL [52] (c) MLP [38] (d) CSF [36] (e) LDT [11] (f) FCN [49] (g) FDN [18] (h) FNBD [39] (i) RGDN [12] (j) SVMAP (ours)…”

Section: Datasets and Implementation Detailsmentioning

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)

View full text Add to dashboard Cite

show abstract

“…Inspired by the artifact removal network (Son and Lee 2017), we propose a de-artifact network A as shown in Figure 3. Different from (Son and Lee 2017), the proposed network introduces the synthesized sketches to encourage the restoration to preserve details and edges. The sharp sketches and the initial deblurred results are concatenated and fed into this network.…”

Section: Network Architecturementioning

confidence: 99%

“…The network is downsampled once to enlarge the receptive field and save the computational cost. Afterward, we use 5 residual blocks which have proven effective in (Son and Lee 2017) for image restoration. The following transposed convolutional layer reconstructs the features with full resolution.…”

Section: Network Architecturementioning

confidence: 99%

Learning to Deblur Face Images via Sketch Synthesis

Lin

Zhang²,

Pan

et al. 2020

AAAI

View full text Add to dashboard Cite

The success of existing face deblurring methods based on deep neural networks is mainly due to the large model capacity. Few algorithms have been specially designed according to the domain knowledge of face images and the physical properties of the deblurring process. In this paper, we propose an effective face deblurring algorithm based on deep convolutional neural networks (CNNs). Motivated by the conventional deblurring process which usually involves the motion blur estimation and the latent clear image restoration, the proposed algorithm first estimates motion blur by a deep CNN and then restores latent clear images with the estimated motion blur. However, estimating motion blur from blurry face images is difficult as the textures of the blurry face images are scarce. As most face images share some common global structures which can be modeled well by sketch information, we propose to learn face sketches by a deep CNN so that the sketches can help the motion blur estimation. With the estimated motion blur, we then develop an effective latent image restoration algorithm based on a deep CNN. Although involving the several components, the proposed algorithm is trained in an end-to-end fashion. We analyze the effectiveness of each component on face image deblurring and show that the proposed algorithm is able to deblur face images with favorable performance against state-of-the-art methods.

show abstract

“…Many works have been done on non-blind image deconvolution using a machine learning approach [14,16,24]. Although they present the proper results about image deblurring, the non-blind approach requires some information about the image a priori and cannot be used in many application domains, for which no information is available about the image a priori.…”

Section: Related Workmentioning

confidence: 99%

A regularized deep learning approach for image de-blurring

Conti

Minucci

Derakhshan

2017

Proceedings of the 1st International Conference on Internet of Things and Machine Learning

View full text Add to dashboard Cite

In this paper, a novel convolutional neural network model for blind deconvolution of images is proposed. The structure of the model is based on two sub models devoted, respectively, to deblurring and denoising of an input image. The model has been designed to restore a picture affected by different kinds of noise. The main innovation is the introduction of a regularization term in the training cost function, based on a blurred/non-blurred classification tool. Results show interesting features of the model, particularly regarding the robustness of results. The comparison with other state-of-the-art models confirms the value of the model proposed in this study.

show abstract

Fast non-blind deconvolution via regularized residual networks with long/short skip-connections

Cited by 40 publications

References 25 publications

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

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

Learning to Deblur Face Images via Sketch Synthesis

A regularized deep learning approach for image de-blurring

Contact Info

Product

Resources

About