Table 8 Comparison of results from the 3 different LipAuth model weights presented in this section, against lighting conditions for the 20 unseen individuals from FAVLIPS
From: Understanding visual lip-based biometric authentication for mobile devices
XM2VTS only | Updated weights | Retrained | |
|---|---|---|---|
XM2VTS: Evaluation Set | 1.21% | 1.95% | 5.60% |
FAVLIPS: Neutral Nums | 22.43% | 13.79% | 10.83% |
FAVLIPS: Light Front | 28.44% | 17.50% | 20.54% |
FAVLIPS: Light Side | 42.29% | 36.67% | 30.00% |
FAVLIPS: Light Behind | 44.91% | 24.17% | 29.12% |