Pose-Invariant Face Recognition via RGB-D Images

Table 2

Recognition rates of different methods on the pose-expression subset of the CurtinFaces database.

Pose	COV+LBP [13]	DCT+SRC [11]			RSRC [14]	Proposed method
		Depth	Texture	Fusion		Depth	Texture	Fusion
Frontal	N/A	100%	100%	100%	100%	100%	100%	100%
±30° yaw	94.2%	88.3%	99.8%	99.4%	99.4%	88.5%	99.5%	99.5%
±60° yaw	84.6%	87.0%	97.4%	98.2%	98.2%	87.7%	97.5%	98.4%
±90° yaw	75.0%	74.0%	83.7%	84.6%	93.5%	78.5%	94.0%	95.1%
±60° pitch	N/A	81.6%	89.1%	92.8%	N/A	82.5%	94.3%	96.7%

For COV+LBP, a subset of pose-variation of the CurtinFaces is used.
For RSRC, we report the average of the best results at the same yaw pose obtained by the method.