Image shadow removal using cycle generative adversarial networks

Shen-Chuan Tai, Peng Yu Chen, Xin An Jiang

Research output: Contribution to journalArticle

Abstract

A shadow-removal algorithm based on cycle generative adversarial network is proposed. There are two networks in the proposed method. To increase the diversity of shadow images for improving the robustness in the shadow-removal process, the first network is used to add shadows in nonshadow images to increase the variation of training data. Then, the second network is trained for the shadow-removal task. Six different losses are calculated and combined in the loss function to increase the performance. Ablation experiments show that the resulting images suffer from some artifacts without any of the six losses in the loss function. The proposed method presents lower value of root-mean-squared error and the superior visual quality compared to the state-of-the-art image removal algorithms.

Original languageEnglish
Article number013034
JournalJournal of Electronic Imaging
Volume28
Issue number1
DOIs
Publication statusPublished - 2019 Jan 1

Fingerprint

cycles
Ablation
ablation
artifacts
education
Experiments

All Science Journal Classification (ASJC) codes

  • Atomic and Molecular Physics, and Optics
  • Computer Science Applications
  • Electrical and Electronic Engineering

Cite this

@article{a96fbcac3fac4615bff74e43f588b0ab,
title = "Image shadow removal using cycle generative adversarial networks",
abstract = "A shadow-removal algorithm based on cycle generative adversarial network is proposed. There are two networks in the proposed method. To increase the diversity of shadow images for improving the robustness in the shadow-removal process, the first network is used to add shadows in nonshadow images to increase the variation of training data. Then, the second network is trained for the shadow-removal task. Six different losses are calculated and combined in the loss function to increase the performance. Ablation experiments show that the resulting images suffer from some artifacts without any of the six losses in the loss function. The proposed method presents lower value of root-mean-squared error and the superior visual quality compared to the state-of-the-art image removal algorithms.",
author = "Shen-Chuan Tai and Chen, {Peng Yu} and Jiang, {Xin An}",
year = "2019",
month = "1",
day = "1",
doi = "10.1117/1.JEI.28.1.013034",
language = "English",
volume = "28",
journal = "Journal of Electronic Imaging",
issn = "1017-9909",
publisher = "SPIE",
number = "1",

}

Image shadow removal using cycle generative adversarial networks. / Tai, Shen-Chuan; Chen, Peng Yu; Jiang, Xin An.

In: Journal of Electronic Imaging, Vol. 28, No. 1, 013034, 01.01.2019.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Image shadow removal using cycle generative adversarial networks

AU - Tai, Shen-Chuan

AU - Chen, Peng Yu

AU - Jiang, Xin An

PY - 2019/1/1

Y1 - 2019/1/1

N2 - A shadow-removal algorithm based on cycle generative adversarial network is proposed. There are two networks in the proposed method. To increase the diversity of shadow images for improving the robustness in the shadow-removal process, the first network is used to add shadows in nonshadow images to increase the variation of training data. Then, the second network is trained for the shadow-removal task. Six different losses are calculated and combined in the loss function to increase the performance. Ablation experiments show that the resulting images suffer from some artifacts without any of the six losses in the loss function. The proposed method presents lower value of root-mean-squared error and the superior visual quality compared to the state-of-the-art image removal algorithms.

AB - A shadow-removal algorithm based on cycle generative adversarial network is proposed. There are two networks in the proposed method. To increase the diversity of shadow images for improving the robustness in the shadow-removal process, the first network is used to add shadows in nonshadow images to increase the variation of training data. Then, the second network is trained for the shadow-removal task. Six different losses are calculated and combined in the loss function to increase the performance. Ablation experiments show that the resulting images suffer from some artifacts without any of the six losses in the loss function. The proposed method presents lower value of root-mean-squared error and the superior visual quality compared to the state-of-the-art image removal algorithms.

UR - http://www.scopus.com/inward/record.url?scp=85062643984&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85062643984&partnerID=8YFLogxK

U2 - 10.1117/1.JEI.28.1.013034

DO - 10.1117/1.JEI.28.1.013034

M3 - Article

VL - 28

JO - Journal of Electronic Imaging

JF - Journal of Electronic Imaging

SN - 1017-9909

IS - 1

M1 - 013034

ER -