TABLE 3.
Performance (%) of the proposed and existing algorithm for video and frame based presentation attack detection on the proposed SnapChat database. The results are reported in terms of the average equal error rate and classification error rates along with standard deviation (±).
| Input | Features | EER | ACER | APCER | BPCER |
|---|---|---|---|---|---|
| Video | LBP (Määttä et al., 2011) | 21.7 ± 6.1 | 21.3 ± 5.9 | 16.51 ± 5.7 | 26.28 ± 6.1 |
| ULBP (Ojala et al., 2002) | 24.5 ± 6.0 | 22.7 ± 5.8 | 12.17 ± 6.0 | 33.45 ± 5.9 | |
| RIULBP Ojala et al. (2002) | 24.7 ± 4.9 | 23.5 ± 4.7 | 16.12 ± 6.2 | 32.62 ± 3.2 | |
| CLBP (Guo et al., 2010) | 24.5 ± 6.1 | 24.8 ± 5.9 | 12.60 ± 5.2 | 37.16 ± 6.6 | |
| Haralick + RDWT (Agarwal et al., 2016) | 25.6 ± 7.2 | 24.5 ± 7.3 | 13.99 ± 8.5 | 35.24 ± 6.1 | |
| BSIF (Kannala and Rahtu, 2012) | 25.2 ± 9.1 | 24.9 ± 9.3 | 29.96 ± 13.0 | 20.07 ± 5.6 | |
| LPQ Ojansivu and Heikkilä (2008) | 22.9 ± 5.2 | 23.9 ± 5.0 | 33.33 ± 5.4 | 14.58 ± 4.6 | |
| Agarwal et al. (2017) | 18.2 ± 5.6 | 18.1 ± 5.5 | 6.71 ± 5.9 | 29.51 ± 5.1 | |
| Proposed (MagNet) | 13.2 ± 3.4 | 12.9 ± 3.2 | 5.62 ± 2.8 | 20.15 ± 3.6 | |
| Frame | LBP (Määttä et al., 2011) | 27.1 ± 4.3 | 27.3 ± 4.1 | 21.83 ± 5.7 | 32.80 ± 2.5 |
| ULBP (Ojala et al., 2002) | 29.0 ± 3.4 | 28.6 ± 3.3 | 17.68 ± 4.2 | 39.70 ± 2.4 | |
| RIULBP (Ojala et al., 2002) | 28.7 ± 3.7 | 28.7 ± 3.9 | 20.94 ± 5.1 | 37.06 ± 2.7 | |
| CLBP (Guo et al., 2010) | 28.7 ± 3.8 | 28.8 ± 3.6 | 18.88 ± 4.7 | 38.80 ± 2.5 | |
| Haralick + RDWT (Agarwal et al., 2016) | 28.9 ± 4.8 | 28.4 ± 4.6 | 21.82 ± 4.3 | 35.08 ± 4.9 | |
| BSIF (Kannala and Rahtu, 2012) | 30.2 ± 7.0 | 30.2 ± 6.9 | 31.45 ± 8.8 | 29.19 ± 5.0 | |
| LPQ (Ojansivu and Heikkilä, 2008) | 28.7 ± 4.0 | 30.4 ± 3.8 | 40.30 ± 3.6 | 20.50 ± 4.0 | |
| Agarwal et al. (2017) | 24.5 ± 5.1 | 25.4 ± 4.9 | 10.60 ± 5.9 | 40.26 ± 3.9 | |
| Proposed (MagNet) | 18.0 ± 0.4 | 17.6 ± 0.3 | 8.72 ± 0.4 | 26.47 ± 0.2 |
Result of the best performing algorithm is highlighted in bold.