An Efficient Robust Eye Localization by Learning the Convolution Distribution Using Eye Template

Table 2

Comparison of accuracies among the state-of-the-art eye localization approaches. Note the robust algorithms with high accuracies on LFPW.

Method	Accuracy				Mean error
	BioID		LFPW		BioID		LFPW
	Right eye	Left eye	Right eye	Left eye	Right eye	Left eye	Right eye	Left eye
Our method	98.1%	98.2%	96.8%	96.9%	2.7%	2.4%	3.4%	3.1%
Deep CNN	99.9%	100%	99.1%	99.4%	1.7%	1.5%	2.1%	2.0%
ASEF	1.2%	0.2%	2.4%	0.6%	121.4%	88%	81.2%	99.2%
nu-SVR	96.1%	95.9%	92.8%	92.8%	4.2%	4.1%	4.9%	4.9%
BORMAN	79.1%	75.8%	78.2%	92.8%	7.1%	7.8%	7.8%	8.8%
CBDS	97.7%	98.9%	87.9%	91.9%	4.1%	3.9%	7.2%	7%
LUXAND	98.9%	98.66%	95.6%	96.8%	4.1%	3.7%	5.6%	4.5%