The Wawo’s latest search engine exhibits considerably enhanced performance over the 2007 HMM-based algorithm. Through tests on FERET database, a publicly available face image database, we observe the correct recognition rate has been improved significantly.

Face Identification

More specifically, the Rank1 Closed Set Identification rates are:

FERET dataset	Rank-1 correct rate
FB (expression)	99,7%
FC (illumination)	100%
Duplication II	91,5
Duplication I	91

Correct recognition rate % at rank 1

BC+40	BH-40	BD+25	BG-25	BE+15	BF-15
85%	77%	100%	99%	99,5%	100%

Face Verification

FRR % on FAR 2%

	Duplication I	Duplication II	FB (expression)	FC (illumination)
Verification 1-1	9,6%	10,7%	0,3%	0,5%
Verification 1-n	0,14%	0,43%	0	0

Performance, rotated images, Face Verification

FRR % at FAR of 2%

BC+40	BH-40	BD+25	BG-25	BE+15	BF-15
4.0%	9.5%	0%	0%	0%	0%

FRR % at FAR of 0.5%

BC+40	BH-40	BD+25	BG-25	BE+15	BF-15
8%	18%	0%	0%	0%	0%

Performance, rotated images, Rank-1 correct identification rate on FERET

About FERET

All measurements were done using the FERET database managed by the Defense Advanced Research Projects Agency (DARPA) and the National Institute of Standards and Technology (NIST). The database contains a total of 14,126 images that includes 1199 individuals and 365 duplicate sets of images. A duplicate set is a second set of images of a person already in the database, usually taken on a different day. The database is organized into a gallery of 1196 frontal face images and four different probe sets. Note that all the tests used a single gallery containing 1196 images. All images show the face in a frontal or nearly frontal pose and were taken in a setting typical for passport photographs. The average resolution of the images is low as the average distance between the eye centers is only 65 pixels.

For more information on the FERET database, please visit Wikipedia.