LBP-based periocular recognition on challenging face datasets

EURASIP Journal on Image and Video Processing

Table 1 Rank-1 match performance on challenging datasets

Datasets (rank-1 accuracy)	Face recognition literature (%)	COTS (%)	Proposed (%)
Georgia Tech DB [12]	96.60 [16]	99.73	92.42
Twins Face DB [13]	96.24 ^a [17]	98.04	98.03
MORPH Album1 [15]	50.4 ^b [18]	51.60	33.20
FRGC [14]	#	99.65	97.51

The table lists the rank-1 accuracies obtained for Georgia Tech face database, Twins face database, MORPH Album 1, and FRGC database against the literature, COTS matcher used in this work, and the proposed algorithm. The number sign indicates that [14] performance measures against the FRGC are based on verification rate of 98 at 0.1% FAR. ^aThe results are from both identical and fraternal twins for [17], while our approach used only the identical twins. ^bAverage of rank-1 accuracies of all age groups with an age gap of 0 to 5 years between the probe and gallery.