Diffusion Model High-Quality Facial Image Detection Algorithm Based on Noise Variation

Zhengyu Zhang; Mingxuan Li; Boxuan Cheng; Yinxuan Qu; Chunyu Li; Yuzhu Yang

doi:10.3788/LOP250542

Laser & Optoelectronics Progress, Volume. 62, Issue 16, 1637001(2025)

Diffusion Model High-Quality Facial Image Detection Algorithm Based on Noise Variation

Zhengyu Zhang², Mingxuan Li^1、*, Boxuan Cheng¹, Yinxuan Qu¹, Chunyu Li¹, and Yuzhu Yang¹

Author Affiliations

¹School of Criminal Investigation, People's Public Security University of China, Beijing 100038, China

²School of Information Network Security, People's Public Security University of China, Beijing 100038, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(10)

Fig. 1. Flowchart of proposed detection model

Download full size

Fig. 2. Flowchart of VAE encoding and decoding

Download full size

Fig. 3. Flowchart of image processing by SE module

Download full size

Fig. 4. Partial images of the self-made high-quality facial image dataset

Download full size

Fig. 5. Hyperparameter tuning experimental results by different methods. (a) SSIM; (b) ORB; (c) Hu Moments Difference

Download full size

Fig. 6. Inversion process and results without image-text multimodal alignment embedding module

Download full size

Table 1. Comparative experimental results of different models

View table

Table 1. Comparative experimental results of different models

Model	Training dataset	Test dataset	Average accuracy
Model	Training dataset	Test dataset	Generated image	Real image
CNNDet^［9］	Original dataset	Genimage*	69.2	57.1
CNNDet^［9］	Self-made dataset	Genimage*	83.4	78.5
F³Net^［11］	Original dataset	Genimage*	50.1	70.6
F³Net^［11］	Self-made dataset	Genimage*	83.6	91.7
DIRE^［13］	Original dataset	Genimage*	81.0	88.3
DIRE^［13］	Self-made dataset	Genimage*	89.0	92.1
CNC （proposed）	Self-made dataset	Genimage*	98.7	100.0

Table 2. Comparative experimental results of different attention mechanisms
View table
Table 2. Comparative experimental results of different attention mechanisms
Attention mechanism F1-score /% Params /10⁶ FLOPs /10⁹ Time /ms
SE 93.4 26.89 4.28 16.1
CBAM 92.8 27.05 4.35 16.8
SK 94.2 28.33 4.57 17.5

Table 3. Ablation experimental results for image-text multimodal alignment embedding module unit: %
View table
Table 3. Ablation experimental results for image-text multimodal alignment embedding module unit: %
Model Training dataset Test dataset Average accuracy
Generated image Realimage
With Self-made dataset Genimage* 97.9 98.1
Without Self-made dataset Genimage** 61.3 69.2

Table 4. Ablation experimental results for SE module
View table
Table 4. Ablation experimental results for SE module
Model F1-score /% Params /10⁶ FLOPs /10⁹ Time /ms
Baseline 91.2 25.56 4.12 15.2
Baseline+SE 93.4 26.89 4.28 16.1

Tools

Get Citation

Copy Citation Text

Zhengyu Zhang, Mingxuan Li, Boxuan Cheng, Yinxuan Qu, Chunyu Li, Yuzhu Yang. Diffusion Model High-Quality Facial Image Detection Algorithm Based on Noise Variation[J]. Laser & Optoelectronics Progress, 2025, 62(16): 1637001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites