WGAN based edge preserved de-blurring using perceptual style similarity

Minsoo Hong, Yoonsik Choe

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Generative Adversarial Network (GAN) is an effective generative model and can be used for de-blurring. In this paper, we propose an edge preserved de-blurring method using Wasserstein generative adversarial network with gradient penalty (WGAN-GP), which is based on conditional GAN. Also, since detailed-edge is the most important factor in de-blurred image, in order to preserve detailed-edge and capture its perceptual similarity, the style loss function is added to represent the perceptual information of the edge. Consequently, the proposed method improves the similarity between sharp images and blurred images by minimizing Wasserstein distance, and well captures the perceptual similarity using style loss function, considering the correlation of features in Convolutional Neural Network. Experiments depict that the proposed method achieves 0.98 in SSIM that is higher performance, compared to other conventional methods such as filter based methods and content based method.

Original languageEnglish
Title of host publicationProceedings - 2018 International Symposium on Computer, Consumer and Control, IS3C 2018
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages58-61
Number of pages4
ISBN (Electronic)9781538670361
DOIs
Publication statusPublished - 2019 Feb 19
Event4th International Symposium on Computer, Consumer and Control, IS3C 2018 - Taichung, Taiwan, Province of China
Duration: 2018 Dec 62018 Dec 8

Publication series

NameProceedings - 2018 International Symposium on Computer, Consumer and Control, IS3C 2018

Conference

Conference4th International Symposium on Computer, Consumer and Control, IS3C 2018
CountryTaiwan, Province of China
CityTaichung
Period18/12/618/12/8

Fingerprint

Deblurring
Loss Function
Neural networks
Wasserstein Distance
Generative Models
Experiments
Penalty
Similarity
Style
High Performance
Neural Networks
Filter
Gradient
Experiment

All Science Journal Classification (ASJC) codes

  • Computer Networks and Communications
  • Control and Systems Engineering
  • Energy Engineering and Power Technology
  • Computer Science Applications
  • Control and Optimization
  • Signal Processing

Cite this

Hong, M., & Choe, Y. (2019). WGAN based edge preserved de-blurring using perceptual style similarity. In Proceedings - 2018 International Symposium on Computer, Consumer and Control, IS3C 2018 (pp. 58-61). [8644758] (Proceedings - 2018 International Symposium on Computer, Consumer and Control, IS3C 2018). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/IS3C.2018.00023
Hong, Minsoo ; Choe, Yoonsik. / WGAN based edge preserved de-blurring using perceptual style similarity. Proceedings - 2018 International Symposium on Computer, Consumer and Control, IS3C 2018. Institute of Electrical and Electronics Engineers Inc., 2019. pp. 58-61 (Proceedings - 2018 International Symposium on Computer, Consumer and Control, IS3C 2018).
@inproceedings{5408684aa2eb46a18d8061f9f1766b9c,
title = "WGAN based edge preserved de-blurring using perceptual style similarity",
abstract = "Generative Adversarial Network (GAN) is an effective generative model and can be used for de-blurring. In this paper, we propose an edge preserved de-blurring method using Wasserstein generative adversarial network with gradient penalty (WGAN-GP), which is based on conditional GAN. Also, since detailed-edge is the most important factor in de-blurred image, in order to preserve detailed-edge and capture its perceptual similarity, the style loss function is added to represent the perceptual information of the edge. Consequently, the proposed method improves the similarity between sharp images and blurred images by minimizing Wasserstein distance, and well captures the perceptual similarity using style loss function, considering the correlation of features in Convolutional Neural Network. Experiments depict that the proposed method achieves 0.98 in SSIM that is higher performance, compared to other conventional methods such as filter based methods and content based method.",
author = "Minsoo Hong and Yoonsik Choe",
year = "2019",
month = "2",
day = "19",
doi = "10.1109/IS3C.2018.00023",
language = "English",
series = "Proceedings - 2018 International Symposium on Computer, Consumer and Control, IS3C 2018",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
pages = "58--61",
booktitle = "Proceedings - 2018 International Symposium on Computer, Consumer and Control, IS3C 2018",
address = "United States",

}

Hong, M & Choe, Y 2019, WGAN based edge preserved de-blurring using perceptual style similarity. in Proceedings - 2018 International Symposium on Computer, Consumer and Control, IS3C 2018., 8644758, Proceedings - 2018 International Symposium on Computer, Consumer and Control, IS3C 2018, Institute of Electrical and Electronics Engineers Inc., pp. 58-61, 4th International Symposium on Computer, Consumer and Control, IS3C 2018, Taichung, Taiwan, Province of China, 18/12/6. https://doi.org/10.1109/IS3C.2018.00023

WGAN based edge preserved de-blurring using perceptual style similarity. / Hong, Minsoo; Choe, Yoonsik.

Proceedings - 2018 International Symposium on Computer, Consumer and Control, IS3C 2018. Institute of Electrical and Electronics Engineers Inc., 2019. p. 58-61 8644758 (Proceedings - 2018 International Symposium on Computer, Consumer and Control, IS3C 2018).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

TY - GEN

T1 - WGAN based edge preserved de-blurring using perceptual style similarity

AU - Hong, Minsoo

AU - Choe, Yoonsik

PY - 2019/2/19

Y1 - 2019/2/19

N2 - Generative Adversarial Network (GAN) is an effective generative model and can be used for de-blurring. In this paper, we propose an edge preserved de-blurring method using Wasserstein generative adversarial network with gradient penalty (WGAN-GP), which is based on conditional GAN. Also, since detailed-edge is the most important factor in de-blurred image, in order to preserve detailed-edge and capture its perceptual similarity, the style loss function is added to represent the perceptual information of the edge. Consequently, the proposed method improves the similarity between sharp images and blurred images by minimizing Wasserstein distance, and well captures the perceptual similarity using style loss function, considering the correlation of features in Convolutional Neural Network. Experiments depict that the proposed method achieves 0.98 in SSIM that is higher performance, compared to other conventional methods such as filter based methods and content based method.

AB - Generative Adversarial Network (GAN) is an effective generative model and can be used for de-blurring. In this paper, we propose an edge preserved de-blurring method using Wasserstein generative adversarial network with gradient penalty (WGAN-GP), which is based on conditional GAN. Also, since detailed-edge is the most important factor in de-blurred image, in order to preserve detailed-edge and capture its perceptual similarity, the style loss function is added to represent the perceptual information of the edge. Consequently, the proposed method improves the similarity between sharp images and blurred images by minimizing Wasserstein distance, and well captures the perceptual similarity using style loss function, considering the correlation of features in Convolutional Neural Network. Experiments depict that the proposed method achieves 0.98 in SSIM that is higher performance, compared to other conventional methods such as filter based methods and content based method.

UR - http://www.scopus.com/inward/record.url?scp=85063211506&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85063211506&partnerID=8YFLogxK

U2 - 10.1109/IS3C.2018.00023

DO - 10.1109/IS3C.2018.00023

M3 - Conference contribution

T3 - Proceedings - 2018 International Symposium on Computer, Consumer and Control, IS3C 2018

SP - 58

EP - 61

BT - Proceedings - 2018 International Symposium on Computer, Consumer and Control, IS3C 2018

PB - Institute of Electrical and Electronics Engineers Inc.

ER -

Hong M, Choe Y. WGAN based edge preserved de-blurring using perceptual style similarity. In Proceedings - 2018 International Symposium on Computer, Consumer and Control, IS3C 2018. Institute of Electrical and Electronics Engineers Inc. 2019. p. 58-61. 8644758. (Proceedings - 2018 International Symposium on Computer, Consumer and Control, IS3C 2018). https://doi.org/10.1109/IS3C.2018.00023