3D Printing Mask Attacks Detection Based on Multi-Feature Fusion

Group	Approach	Gallery set	Probe set
Group	Approach	Gallery set	Genuine face	Fraud mask
Base-1Base-2Base-3Base-4	3D keypoint matching3D keypoint matching3D keypoint matchingBottleneck feature fusion	$\begin{matrix} T_{B}^{1} \end{matrix}$ $\begin{matrix} T_{F}^{1} \end{matrix}$ + $\begin{matrix} T_{B}^{1} \end{matrix}$ $\begin{matrix} T_{F}^{1} \end{matrix}$ + $\begin{matrix} T_{B}^{1} \end{matrix}$ $\begin{matrix} T_{F}^{1} \end{matrix}$ + $\begin{matrix} T_{B}^{1} \end{matrix}$	$\begin{matrix} T_{B}^{i} \end{matrix}$ $\begin{matrix} T_{F}^{i} \end{matrix}$ + $\begin{matrix} T_{B}^{i} \end{matrix}$ $\begin{matrix} T_{F}^{i} \end{matrix}$ + $\begin{matrix} T_{B}^{i} \end{matrix}$ $\begin{matrix} T_{F}^{i} \end{matrix}$ + $\begin{matrix} T_{B}^{i} \end{matrix}$	--F_BF_B
Anti-1Anti-2Anti-3Anti-4	Method in Ref.[15]Method in Ref. [10]Method in Ref. [16]Raw feature fusion	$\begin{matrix} T_{B}^{1} \end{matrix}$ $\begin{matrix} T_{B}^{1} \end{matrix}$ $\begin{matrix} T_{B}^{1} \end{matrix}$ $\begin{matrix} T_{B}^{1} \end{matrix}$	----	F_BF_BF_BF_B
Anti-5Anti-6	Score fusionBottleneck feature fusion	$\begin{matrix} T_{B}^{1} \end{matrix}$ $\begin{matrix} T_{B}^{1} \end{matrix}$	--	F_BF_B

Table 2. Experimental results of verification and anti-spoofing performance evaluation%
View table
Table 2. Experimental results of verification and anti-spoofing performance evaluation%
R_FAR Base-1 TAR Base-2 TAR Base-3 STRR Base-4 STRR Base-4 HTER
0.1 95.1 - 44.3 96.7 7.4
0.05 93.4 97.6 50.4 - -
0.01 92.6 94.6 62.6 - -
0.001 - 90.9 70.4 - -

Table 3. Anti-attack spoofing performance comparison of different algorithms

View table

Table 3. Anti-attack spoofing performance comparison of different algorithms

Group	Approach	HTER
Anti-1Anti-2Anti-3	Method in Ref.[15]Method in Ref. [10]Method in Ref. [16]	22.416.212.5
Anti-4	Raw feature SVM	16.5
Anti-5	Raw feature fusion	8.8
Anti-6Anti-7	Score fusionBottleneck feature fusion	15.34.7

Table 4. Comparison between training time and testing time of different methodss
View table
Table 4. Comparison between training time and testing time of different methodss
Approach Training time Test time
Multi-layer perceptronBottleneck feature fusion 2470.7147.5 0.2580.024

Tools

Get Citation

Copy Citation Text

Jingwei Lu, Hetian Chen, Xiaopan Ma, Jimin Chen. 3D Printing Mask Attacks Detection Based on Multi-Feature Fusion[J]. Laser & Optoelectronics Progress, 2019, 56(3): 031002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Image Processing

Received: Jul. 4, 2018

Accepted: Aug. 13, 2018

Published Online: Jul. 31, 2019

The Author Email: Jingwei Lu (18810815230@126.com)

DOI:10.3788/LOP56.031002

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Experimental setup parameters

Table 1. Experimental setup parameters

Table 2. Experimental results of verification and anti-spoofing performance evaluation%

Table 2. Experimental results of verification and anti-spoofing performance evaluation%

Table 3. Anti-attack spoofing performance comparison of different algorithms

Table 3. Anti-attack spoofing performance comparison of different algorithms

Table 4. Comparison between training time and testing time of different methodss

Table 4. Comparison between training time and testing time of different methodss