Table 2.

Classification accuracy on the FERG dataset.

Method	Accuracy Rate
DeepExpr [2]	89.02%
Ensemble Multi-feature [53]	97%
Adversarial NN [54]	98.2%
The proposed algorithm	99.3%