Laser & Optoelectronics Progress, Volume. 62, Issue 8, 0837002(2025)
Three-Dimensional Facial UV Texture Restoration Based on Gated Convolution
Figures & Tables(9)
Fig. 1. Acquisition of UV texture map of human face. (a) Input; (b) 3D facial model; (c) UV texture
Fig. 4. UV texture image restoration effect. (a) Original images; (b) incomplete UV texture; (c) inpainting UV texture
Fig. 5. Comparison of UV texture completion effects using different algorithms. (a) Front face; (b) origin; (c) uv texture of frontal face; (d) input; (e) PRNet; (f) 3DDFA; (g) proposed algorithm
Fig. 6. Comparison of 3D models using different algorithms. (a) Original texture; (b) PRNet; (c) 3DDFA; (d) proposed algorithm
Fig. 7. Comparison of repair results with and without spectral normalization loss
Table 1. Comparison results of PSNR and SSIM indicators using different methods
View tableTable 1. Comparison results of PSNR and SSIM indicators using different methods
Index Algorithm Viewing angle 20°‒30° 40°‒50° 70°‒80° PSNR /dB PRNet 28.72 28.41 27.99 3DDFA 28.45 28.11 27.67 Proposed algorithm 29.06 28.75 28.16 SSIM PRNet 0.66 0.61 0.51 3DDFA 0.55 0.50 0.40 Proposed algorithm 0.77 0.74 0.63
Table 2. Comparison results of PSNR and SSIM indicators with and without spectral normalization loss
View tableTable 2. Comparison results of PSNR and SSIM indicators with and without spectral normalization loss
Loss function PSNR /dB SSIM loss 28.40 0.61 loss 28.66 0.71
Tools
Get Citation
Copy Citation Text
Zhenhua Zhao, Bo Yang, Qiuhang Chen, Yiming Ying. Three-Dimensional Facial UV Texture Restoration Based on Gated Convolution[J]. Laser & Optoelectronics Progress, 2025, 62(8): 0837002
Paper Information
Category: Digital Image Processing
Received: Aug. 13, 2024
Accepted: Sep. 24, 2024
Published Online: Apr. 2, 2025
The Author Email: Bo Yang (1351939582@qq.com)
CSTR:32186.14.LOP241833
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20