Skip to main content

Table 5 Experimented face landmarking algorithms

From: A comparative study of face landmarking techniques

Acronym

Face detector

#Landmarks

Training set

Face pose

Processing time

    

and expression

per image

FFPD [44]

Haar feature based

20

CK frontal faces

Frontal; Neutral

0.85

 

GentleBoost classifier

    

AAM [125]

Viola-Jones face

66

Multi-PIE, XM2VTS

Near-frontal;

0.12 s

 

detector

  

Expression

 

STASM [31]

Viola-Jones and Rowley

76

XM2VTS, AR

Near-frontal;

0.18 s

 

face detector [126]

  

Expression

 

BORMAN [70]

Viola-Jones face detector

22

FERET, MMI

Near-frontal;

65 s

 

detector

  

Expression

 

ZhuRamanan [99]

A mixture of tree

68

Multi-PIE

Free of pose and

25 s

 

structured part models

  

expression

 

Everingham [102]

Viola-Jones face detector

9

Consumer images

Near frontal

0.4 s

flandmark [103]

Description NA

7

LFW

Near frontal;

0.12 s

    

expression

 
  1. Average run time on BioID database with a CPU of 2.50GHz and 8GB RAM. Each image has a resolution of 384×286.
  2. Trained model with 1050 parts.