Research Article
Research on Multimodality Face Antispoofing Model Based on Adversarial Attacks
Table 5
The accuracy of the model before and after adding the defence module to the IR 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 | 96.3 | 95.2 | 77 | 91.9 | | FGSM 0.01 | 1 | 57.3 | 66 | 79.1 | | DeepFool | 10.7 | 14.5 | 67.6 | 82.3 | | NATTACK | 12.4 | 46.3 | 78.1 | 89.6 |
|
|
