Research on Multimodality Face Antispoofing Model Based on Adversarial Attacks

Table 4

The accuracy of the model before and after adding the defence module to the Depth 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	98.5	88.3	93.65	90.3
FGSM 0.01	63.9	70.2	82.3	89.7
DeepFool	4.9	50.8	64.5	76.6
NATTACK	7.6	55.98	91.2	84.5