Research on Multimodality Face Antispoofing Model Based on Adversarial Attacks

Table 2

The accuracy of multimodality model after resisting the single-modality adversarial attacks.

Attack method	Attack RGB (%)	Attack Depth (%)	Attack IR (%)
Original sample	99.5	99.5	99.5
FGSM 0.01	99.2	99.1	99.2
FGSM 0.1	99	88.4	98.1
FGSM 0.2	46.4	58.3	83.2
FGSM 0.25	33.3	37.1	65.7
DeepFool	99.3	99.1	99.2
NATTACK	97.1	93.7	97.2