Research Article
Research on Multimodality Face Antispoofing Model Based on Adversarial Attacks
Table 3
The accuracy of the model before and after adding the defence module to the RGB images.
| | | The original authors’ pretrained model (%) | After adding the diffusible high-frequency suppression (%) | After using the mixed adversarial training (%) | Combination of the high-frequency suppression module and the mixed adversarial training (%) |
| | Original | 94.1 | 94.4 | 82.3 | 92.7 | | FGSM 0.01 | 2.5 | 8.9 | 70.8 | 81.5 | | DeepFool | 8 | 60.3 | 72.6 | 81 | | NATTACK | 24.3 | 76.6 | 63.25 | 85.96 |
|
|
