Applications - 3D Face Modeling, Analysis and Recognition - page 148

Table 5.1 3D face data sets for 3D face recognition

Number of Total Missing

Database Sensor subjects scans data Occlusions

FRGC v1.0 laser 275 943 No No

FRGC v2.0 laser 466 4007 Yes No

GAVAB laser 61 549 Yes Yes

Bosphorus structured light 105 4652 Yes Yes

In Section 5.4, state-of-the-art methods for facial expression recognition are reviewed,
and two specific solutions are then discussed in more detail. A semi-automatic approach
that requires manual intervention, and a fully automatic solution that performs expression
recognition without any human intervention. In this section, the new challenges posed by the
analysis of 3D dynamic face sequences for the purpose of facial expression recognition are
also addressed, and a recent method giving effective solution to this problem is presented.

•

5.2 3D Face Databases
In the last few years, several 3D face databases have been made publicly available for test-
ing algorithms that use 3D data to perform face modeling, analysis, and recognition. These
databases have progressively included face scans of subjects that exhibit non-neutral facial
expressions and nonfrontal poses. Table 5.1, reports the more general data sets that are cur-
rently available for the task of 3D face recognition. For each data set information about the
sensor used during acquisition, the total number of involved subjects and scans are reported,
together with notes about the presence of scans with missing parts or occlusions.
Table 5.2 summarizes the characteristics of some of the most known and used 3D face
databases that include subjects with non-neutral expressions.
Some data sets have specific characteristics to perform facial expression classification (see
Table 5.3). In this case, it is relevant that the number of expressions labeled in the data set and
that constitute the ground truth for the purpose of expression recognition. Among these data
sets, those constructed at the Binghamton University (BU-3DFE database) (Yin et al., 2006)

Table 5.2 Main characteristics of the most used 3D face databases that include non-neutral facial

acquisitions

Data set Subjects Scans Expressions Pose

FRGC v2.0 466 4007 not categorized - disgust, happiness

sadness, surprise

small changes

GAVAB 61 549 smile, laugh, random up, down, left, right

BU-3DFE 100 2500 anger, disgust, fear happiness,

sadness, surprise

frontal

Bosphorus 105 4666 action units, anger, disgust fear,

happiness, sadness, surprise

13 yaw and pitch rotations
hand, eyeglasses