Image Shadow Removal Based on Minimum Noise Fraction and Generative Adversarial Network

Dong Ding; Jiali Wang; Ming Chen

doi:10.3788/LOP213421

Laser & Optoelectronics Progress, Volume. 60, Issue 4, 0428004(2023)

Image Shadow Removal Based on Minimum Noise Fraction and Generative Adversarial Network

Dong Ding^{1,1,2,2、">">}, Jiali Wang^1,1、">, and Ming Chen^1,1,2、">*

Author Affiliations

¹College of Information Technology, Shanghai Ocean University, Shanghai 201306, China

²Key Laboratory of Fisheries Information, Ministry of Agriculture, Shanghai 201306, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(13)

Fig. 1. MNF results. (a) Original image; (b) the first three features of MNF image (displayed as R, G, B); (c) the first feature of MNF image; (d) the second feature of MNF image; (e) the third feature of MNF image

Download full size

Fig. 2. Training flow of MNF-GAN algorithm

Download full size

Fig. 3. ST-CGAN structure

Download full size

Fig. 4. Generative network structure

Download full size

Fig. 5. Discriminative network structure

Download full size

Fig. 6. Comparison of shadow removal effect between ST-CGAN and MNF-GAN(first three features). (a) Original shadow image; (b) shadow-free image (reference image); (c) ST-CGAN; (d) MNF-GAN (first three features)

Download full size

Fig. 7. Comparison of shadow removal effects of different MNF-GANs. (a) Original shadow image; (b) shadow-free image (reference image); (c) MNF-GAN (the first feature); (d) MNF-GAN (the second feature); (e) MNF-GAN (the third feature); (f) MNF-GAN (first three features)

Download full size

Fig. 8. Comparison of shadow removal effects of different methods on SRD dataset. (a) Original shadow image; (b) shadow-free image (reference image); (c) DeshadowNet; (d) Mask-ShadowGAN; (e) CGAN; (f) proposed algorithm

Download full size

Table 1. Comparison between ISTD dataset and other mainstream shadow datasets

View table

Table 1. Comparison between ISTD dataset and other mainstream shadow datasets

Dataset	Data volume	Content of images	Type	Usage
SBU	4727	Shadow/shadow mask	Pair	Detection
UCF	245	Shadow/shadow mask	Pair	Detection
SRD	3088	Shadow/shadow-free	Pair	Removal
UIUC	76	Shadow/shadow-free	Pair	Removal
LRSS	37	Shadow/shadow-free	Pair	Removal
ISTD	1870	Shadow/shadow mask/shadow-free	Triplet	Both

Table 2. Statistical results of shadow removal between ST-CGAN and MNF-GAN (first three features)
View table
Table 2. Statistical results of shadow removal between ST-CGAN and MNF-GAN (first three features)
Algorithm SSIM RMSE
ST-CGAN 0.9401 15.6675
MNF-GAN（first three features） 0.9494 13.5222

Table 3. Comparison of shadow removal effects between MNF-GAN based on different features and ST-CGAN

View table

Table 3. Comparison of shadow removal effects between MNF-GAN based on different features and ST-CGAN

Algorithm	SSIM	RMSE	Rising rate of SSIM/%	Decline rate of RMSE/%
ST-CGAN	0.9401	15.6675
MNF-GAN（the first feature）	0.9478	14.1621	0.82	9.61
MNF-GAN（the second feature）	0.9510	12.3016	1.16	21.48
MNF-GAN（the third feature）	0.9499	13.8513	1.04	11.59
MNF-GAN（first three features）	0.9494	13.5222	0.99	13.69

Table 4. Statistical results of shadow removal batween ST-CGAN and PCA-GAN and MNF-GAN (the second feature)
View table
Table 4. Statistical results of shadow removal batween ST-CGAN and PCA-GAN and MNF-GAN (the second feature)
Algorithm SSIM RMSE
ST-CGAN 0.9401 15.6675
PCA-GAN 0.9450 14.9006
MNF-GAN（the second feature） 0.9510 12.3016

Table 5. Shadow removal quantitative comparison of different algorithms on SRD dataset
View table
Table 5. Shadow removal quantitative comparison of different algorithms on SRD dataset
Parameter DeshadowNet Mask-ShadowGAN CGAN Proposed algorithm
SSIM 0.9328 0.9465 0.9550 0.9780
RMSE 13.5324 14.7500 12.9437 9.8717

Tools

Get Citation

Copy Citation Text

Dong Ding, Jiali Wang, Ming Chen. Image Shadow Removal Based on Minimum Noise Fraction and Generative Adversarial Network[J]. Laser & Optoelectronics Progress, 2023, 60(4): 0428004

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Remote Sensing and Sensors

Received: Dec. 31, 2021

Accepted: Mar. 29, 2022

Published Online: Feb. 14, 2023

The Author Email: Chen Ming (mchen@shou.edu.cn)

DOI:10.3788/LOP213421

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Comparison between ISTD dataset and other mainstream shadow datasets

Table 1. Comparison between ISTD dataset and other mainstream shadow datasets

Table 2. Statistical results of shadow removal between ST-CGAN and MNF-GAN (first three features)

Table 2. Statistical results of shadow removal between ST-CGAN and MNF-GAN (first three features)

Table 3. Comparison of shadow removal effects between MNF-GAN based on different features and ST-CGAN

Table 3. Comparison of shadow removal effects between MNF-GAN based on different features and ST-CGAN

Table 4. Statistical results of shadow removal batween ST-CGAN and PCA-GAN and MNF-GAN (the second feature)

Table 4. Statistical results of shadow removal batween ST-CGAN and PCA-GAN and MNF-GAN (the second feature)

Table 5. Shadow removal quantitative comparison of different algorithms on SRD dataset

Table 5. Shadow removal quantitative comparison of different algorithms on SRD dataset