Graphics Reference
In-Depth Information
3D acquisition
2D acquisition
Figure 5.1
FRGC database: 3D and 2D sample data for the subject with
id
=
02463 are reported. It
can be observed as the 3D and 2D acquisitions are registered and both provided with a set of 480
×
640
points/pixels
and 3D acquisitions). Figure 5.1 shows the 3D and 2D data for the gallery scan of a sample
subject. The database can be summarized as follows:
Pros
—largest number of subjects and very high number of scans; textured images; most
used for competitive evaluation; used for the FRGC contest in 2005/2006
•
Cons
—small/moderate facial expressions; expressions not categorized; some images have
small misalignment between 3D and 2D data
•
The GAVAB Database (GAVAB)
The Grupo de Algorıtmica para la Visi on Artificial y la Biometrıa database (GAVAB DB)
(Moreno and Sanchez, 2004a) is characterized by facial scans with very large pose and expres-
sion variations and acquisition. It includes 3D face scans of 61 adult Caucasian individuals (45
males and 16 females). For each individual, nine scans are taken that differ in the acquisition
viewpoint and facial expressions, resulting in a total of 549 facial scans. In particular, for each
individual, there are two frontal face scans with
neutral
expression, two face scans where
the subject is acquired with a
rotated
posture of the face (around
35
◦
looking up or looking
down) and neutral facial expression, and three frontal scans in which the person
laughs
,
smiles
,
or shows a
random
expression. Finally, there are also a right and a left side scan each nominally
acquired with a rotation of
±
90
◦
left and right. This results in about 67% of the scans that
have a neutral expression, but just 22% that have neutral expression and frontal pose.
Modified scans of this database have been used as data for the SHREC 2008
Shape Retrieval
Contest of 3D Face Scans
(Daoudi et al., 2008) and to test face recognition accuracy as well
as to test recognition performance in the case parts of the face scans are missing (Drira et al.,
2010; Huang et al., 2012). As an example, Figure 5.2 shows one 3D neutral scan and the left
and right scans of a sample subject. In summary, the main features of the database are:
±
Pros
—accentuated facial expressions; pose variations (left/right, up/down)
•
Cons
—few subjects (61); small number of scans (549); large artifacts and noisy acquisitions;
no-texture images
•
