Face Liveness Detection Algorithm Based on Real Face Category Adversarial Mechanism

Lei Zhang; Shaoyan Gai; Feipeng Da

doi:10.3788/LOP220649

Journals >Laser & Optoelectronics Progress >Volume 60 >Issue 10 >Page 1010007 > Article

Laser & Optoelectronics Progress
Vol. 60, Issue 10, 1010007 (2023)

Face Liveness Detection Algorithm Based on Real Face Category Adversarial Mechanism

Lei Zhang^1,2, Shaoyan Gai^1,2,*, and Feipeng Da^1,2,3

Author Affiliations

¹School of Automation, Southeast University, Nanjing 210096, Jiangsu , China

²Key Laboratory of Measurement and Control of Complex Systems of Engineering, Ministry of Education, Southeast University, Nanjing 210096, Jiangsu , China

³Shenzhen Research Institute, Southeast University, Shenzhen 518063, Guangdong , China

show less

DOI: 10.3788/LOP220649 Cite this Article Set citation alerts

Lei Zhang, Shaoyan Gai, Feipeng Da. Face Liveness Detection Algorithm Based on Real Face Category Adversarial Mechanism[J]. Laser & Optoelectronics Progress, 2023, 60(10): 1010007 Copy Citation Text

show less

Fig. 1. Motivation of different methods

Download full size

Fig. 2. Overall architecture of the proposed model

Download full size

Fig. 3. Multi-scale attention fusion module

Download full size

Fig. 4. Illustration of triplet loss

Download full size

Fig. 5. t-SNE visualization for the classification features obtained by baseline method and proposed method under the I&C&M to O testing task

Download full size

Fig. 6. Grad-CAM visualization under the I&C&M to O testing task

Download full size

Dataset	Number of identities	Number of videos	Number of real categories	Number of fake categories
MSU-MFSD	35	280	2	6
CASIA-FASD	50	600	3	9
Idiap Replay-Attack	50	1200	4	20
OULU-NPU	55	4950	18	72

Table 1. Details of four datasets

Method	O&C&I to M		O&M&I to C		O&C&M to I		I&C&M to O
Method	HTER /%	AUC /%	HTER /%	AUC /%	HTER /%	AUC /%	HTER /%	AUC /%
w /o att	5.27	95.20	12.11	94.31	10.07	95.99	13.54	93.22
w /o ad	5.71	96.10	10.77	94.51	14.92	93.05	12.30	94.97
w /o tri	7.14	96.53	10.66	95.19	21.42	80.51	22.06	86.37
all	4.52	97.24	9.88	95.46	9.21	96.97	11.45	95.32

Table 2. Evaluation results of different components of the proposed method

Method	O&C&I to M		O&M&I to C		O&C&M to I		I&C&M to O
Method	HTER /%	AUC /%	HTER /%	AUC /%	HTER /%	AUC /%	HTER /%	AUC /%
SSDG	7.38	97.17	10.44	95.94	11.71	96.59	15.61	91.54
Proposed method	4.52	97.24	9.88	95.46	9.21	96.97	11.45	95.32

Table 3. Comparison results between the proposed method and the corresponding baseline method

Method	M&I to C		M&I to O
Method	HTER /%	AUC /%	HTER /%	AUC /%
MS-LBP^［23］	51.16	52.09	43.63	58.07
IDA^［4］	45.16	58.80	54.52	42.17
LBP-TOP^［24］	45.27	54.88	47.26	50.21
MADDG^［10］	41.02	64.33	39.35	65.10
SSDG-M^［9］	31.89	71.29	36.01	66.88
DRDG^［11］	31.28	71.50	33.35	69.14
DASN^［12］	21.48	83.41	21.74	80.87
Proposed method	17.88	89.91	22.70	86.17

Table 4. Comparison result of different domain generalization methods in the case of limited source domains

Backbone	FLOPs /10⁹	Params /10⁶	Speed /（frame·s^-1）	Avg HTER /%	Avg AUC /%
Resnet50	5.37	25.60	46.57	12.05	93.51
Resnet34	4.79	21.81	52.65	12.31	93.87
Resnet18	2.37	11.70	100.75	8.76	96.29

Table 5. Comparison result of different backbone networks

Method	O&C&I to M		O&M&I to C		O&C&M to I		I&C&M to O
Method	HTER /%	AUC /%	HTER /%	AUC /%	HTER /%	AUC /%	HTER /%	AUC /%
MS-LBP^［23］	29.76	78.50	54.28	44.98	50.30	51.64	50.29	49.31
Binary CNN^［25］	29.25	82.87	34.88	71.94	34.47	65.88	29.61	77.54
IDA^［4］	66.67	27.86	55.17	39.05	28.35	78.25	54.20	44.59
Color Texture^［26］	28.09	78.47	30.58	76.89	40.40	62.78	63.59	32.71
LBP-TOP^［24］	36.90	70.80	33.52	73.15	29.14	71.69	30.17	77.61
Auxiliary（Depth）	22.72	85.88	33.52	73.15	29.14	71.69	30.17	77.61
Auxiliary^［27］	—	—	28.40	—	27.60	—	—	—
MADDG^［10］	17.69	88.06	24.50	84.51	22.19	84.99	27.89	80.02
SSDG^［9］	7.38	97.17	10.44	95.94	11.71	96.59	15.61	91.54
DRDG^［11］	12.43	95.81	19.05	88.79	15.56	91.79	15.63	91.75
DASN^［12］	8.33	96.31	12.04	95.33	13.38	86.63	11.77	94.65
Proposed method	4.52	97.24	9.88	95.46	9.21	96.97	11.45	95.32

Table 6. Comparison result between the proposed method and other methods for domain generalization on face anti-spoofing

Lei Zhang, Shaoyan Gai, Feipeng Da. Face Liveness Detection Algorithm Based on Real Face Category Adversarial Mechanism[J]. Laser & Optoelectronics Progress, 2023, 60(10): 1010007

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information

微信扫一扫：分享

微信扫一扫：分享