Table 8.
Detection errors (ACERs) of various face-PAD methods using a subset of the CASIA dataset according to the quality and type of presentation attack samples (unit: %).
| Detection Method | Quality of the Presentation Attack Samples | Type of Presentation Attack Samples | ||||
|---|---|---|---|---|---|---|
| Low Quality Dataset | Normal Quality Dataset | High Quality Dataset | Wrap-Photo Dataset | Cut-Photo Dataset | Video Access Dataset | |
| Baseline Method [13] | 13.0 | 13.0 | 26.0 | 16.0 | 6.0 | 24.0 |
| IQA [18] | 31.7 | 22.2 | 5.6 | 26.1 | 18.3 | 34.4 |
| LBP-TOP [24] | 10.0 | 12.0 | 13.0 | 6.0 | 12.0 | 10.0 |
| LBP + Fisher Score + SVM [22] | 7.2 | 8.8 | 14.4 | 12.0 | 10.0 | 14.7 |
| Patch-based Classification [18] | 5.26 | 6.00 | 5.30 | 5.78 | 5.49 | 5.02 |
| LBP of Color Texture Image [19] | 7.8 | 10.1 | 6.4 | 7.5 | 5.4 | 8.4 |
| CNN + MLBP [27] | 1.834 | 3.950 | 2.210 | 2.054 | 0.545 | 4.835 |
| Proposed Method (FLF) | 2.096 | 3.354 | 1.484 | 1.886 | 0.425 | 1.611 |
| Proposed method (SLF) | 1.417 | 0.040 | 1.085 | 2.005 | 0.428 | 1.423 |