Research on Embroidery Image Restoration Based on Improved Deep Convolutional Generative Adversarial Network

Yixuan Liu; Guangying Ge; Zhenling Qi; Zhenxuan Li; Fulin Sun

doi:10.3788/LOP223060

Laser & Optoelectronics Progress, Volume. 60, Issue 20, 2010005(2023)

Research on Embroidery Image Restoration Based on Improved Deep Convolutional Generative Adversarial Network

Yixuan Liu¹, Guangying Ge^2、*, Zhenling Qi¹, Zhenxuan Li¹, and Fulin Sun¹

¹Shandong Provincial Key Laboratory of Optical Communication Science and Technology, School of Physical Sciences and Information Engineering, Liaocheng University, Liaocheng 252059, Shandong , China

²School of Computer Science and Technology, Liaocheng University, Liaocheng 252059, Shandong , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(14)

Fig. 1. Generative adversarial network

Download full size

Fig. 2. Network structure of DCGAN

Download full size

Fig. 3. CBAM structure. (a) Overall structure of CBAM; (b) CAM structure in CBAM; (c) SAM structure in CBAM

Download full size

Fig. 4. Dilated convolution diagram. (a) Void rate is 1; (b) void rate is 2; (c) void rate is 4

Download full size

Fig. 5. Improved DCGAN image inpainting model

Download full size

Fig. 6. Improved generator model structure of DCGAN

Download full size

Fig. 7. Improved discriminator model structure of DCGAN

Download full size

Fig. 8. Repair process of Egdeconnect algorithm

Download full size

Fig. 9. Comparison chart of repair effect of each method

Download full size

Fig. 10. Comparison chart of repair effect before and after algorithm improvement

Download full size

Fig. 11. heat map matrix and heat map before and after algorithm improvement

Download full size

Table 1. Partial values of $L_{M S E}$ and $L_{a d v}$
View table
Table 1. Partial values of $L_{M S E}$ and $L_{a d v}$
Epoch /10⁴ $L_{M S E}$ $L_{a d v}$ （G） $L_{a d v}$ （D）
0 1.506 0.793 1.227
10 2.063 0.439 0.490
20 2.822 0.372 0.322
30 2.278 0.273 0.286
40 2.165 0.157 0.329

Table 2. Objective evaluation scores of different algorithms for repairing regular rectangular defect areas

View table

Table 2. Objective evaluation scores of different algorithms for repairing regular rectangular defect areas

Algorithm	PSNR /dB	SSIM	Parameter quantity /MB	FLOPs /GB
CE	19.3027	0.7187	68.35	16.69
GL	19.8657	0.7436	66.07	53.11
GConv	21.0095	0.7806	100.33	63.59
EdgeConnect	20.8723	0.7859	103.75	91.03
DCGAN	20.2844	0.6883	88.67	28.74
Proposed method	21.3007	0.7863	96.55	34.40

Table 3. Objective evaluation score of random defect area before and after algorithm improvement
View table
Table 3. Objective evaluation score of random defect area before and after algorithm improvement
Algorithm PSNR /dB SSIM Parameter quantity /MB FLOPs /GB
No improvement measures 20.55680 0.87229 96.55 34.40
Dilated convolution 22.70757 0.92859 98.23 52.27
Dilated convolution+CBMA 24.18276 0.95521 98.75 52.89

Tools

Get Citation

Copy Citation Text

Yixuan Liu, Guangying Ge, Zhenling Qi, Zhenxuan Li, Fulin Sun. Research on Embroidery Image Restoration Based on Improved Deep Convolutional Generative Adversarial Network[J]. Laser & Optoelectronics Progress, 2023, 60(20): 2010005

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Image Processing

Received: Nov. 15, 2022

Accepted: Jan. 6, 2023

Published Online: Sep. 28, 2023

The Author Email: Guangying Ge (ggysd@126.com)

DOI:10.3788/LOP223060

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Partial values of LMSE and Ladv

Table 1. Partial values of LMSE and Ladv

Table 2. Objective evaluation scores of different algorithms for repairing regular rectangular defect areas

Table 2. Objective evaluation scores of different algorithms for repairing regular rectangular defect areas

Table 3. Objective evaluation score of random defect area before and after algorithm improvement

Table 3. Objective evaluation score of random defect area before and after algorithm improvement

Table 1. Partial values of $L_{M S E}$ and $L_{a d v}$

Table 1. Partial values of $L_{M S E}$ and $L_{a d v}$