Face Spoof Attack Recognition Using Discriminative Image Patches

Table 3

Comparison of the proposed method (with SVM, QDA, Naive-Bayes (NB), and Ensemble based classifiers) on REPLAY-ATTACK database with existing methods.

Method	HTER (%)
Multi-LBP [23]	20.25
IQA [3]	15.20
GLCM (Unicamp) [24]	15.62
[4]	16.10
[4]	13.87
PCA + LBP + SVM [25]	20.50
Motion [16]	11.70
DoG-LBP + SVM [1]	11.10
LBP-TOP [26]	8.51
IDA [1]	7.41
Proposed: DF-SVM	6.87
Proposed: DF-NB	8.01
Proposed: DF-QDA	7.30
Proposed: DF-Ensemble	6.23
Proposed: CS-SVM	6.25
Proposed: CS-NB	7.44
Proposed: CS-QDA	6.87
Proposed: CS-Ensemble	6.00
Proposed: DEND-CLUSTERING-SVM	5.98
Proposed: DEND-CLUSTERING-NB	8.87
Proposed: DEND-CLUSTERING-QDA	6.11
Proposed: DEND-CLUSTERING-Ensemble	5.00
Proposed: IQA-SVM	6.23
Proposed: IQA-NB	11.05
Proposed: IQA-QDA	7.75
Proposed: IQA-Ensemble	5.62
Proposed: IPI-SVM	7.50
Proposed: IPI-NB	8.30
Proposed: IPI-QDA	6.19
Proposed: IPI-Ensemble	6.00
Proposed: CP-SVM	8.37
Proposed: CP-NB	9.18
Proposed: CP-QDA	7.12
Proposed: CP-Ensemble	6.80
Proposed: MAXDIST-SVM	5.87
Proposed: MAXDIST-NB	8.01
Proposed: MAXDIST-QDA	6.12
Proposed: MAXDIST-Ensemble	5.00