Paper Info
Title
A Landmark Paper in Face Recognition
Abstract
Good registration (alignment to a reference) is es- sential for accurate face recognition. The effects of the number of landmarks on the mean localization er- ror and the recognition performance are studied. Two landmarking methods are explored and compared for that purpose: (1) the Most Likely-Landmark Locator (MLLL), based on maximizing the likelihood ratio (2), and (2) Viola-Jones detection (14). Both use the loca- tions of facial features (eyes, nose, mouth, etc) as land- marks. Further, a landmark-correction method (BILBO) based on projection into a subspace is introduced. The MLLL has been trained for locating 17 land- marks and the Viola-Jones method for 5. The mean lo- calization errors and effects on the verification perfor- mance have been measured. It was found that on the eyes, the Viola-Jones detector is about 1% of the inter- ocular distance more accurate than the MLLL-BILBO combination. On the nose and mouth, the MLLL-BILBO combination is about 0.5% of the inter-ocular distance more accurate than the Viola-Jones detector. Using more landmarks will result in lower equal-error rates, even when the landmarking is not so accurate. If the same landmarks are used, the most accurate landmark- ing method will give the best verification performance.
Year
DOI
Venue
2006
10.1109/FGR.2006.10
FG
Keywords
Field
DocType
face recognition,landmark paper,landmark cor- rection,face registration,likelihood ratio,viola-jones,land- marking,image registration,face detection,viola jones,detectors,probability density function,shape
Facial recognition system,Computer vision,Viola–Jones object detection framework,Face registration,Subspace topology,Pattern recognition,Computer science,Artificial intelligence,Face detection,Landmark,Detector,Image registration
Conference
ISBN
Citations 
PageRank 
0-7695-2503-2
15
0.75
References 
Authors
5
4
Name
Order
Citations
PageRank
G. M. Beumer1302.09
Q. Tao2150.75
A. M. Bazen3433.45
R. N. J. Veldhuis4887.52