Graphics Reference
In-Depth Information
non-textured scans are reported). With respect to the other data sets, the characteristics of the
Bosphorus database can be summarized as:
Pros
—categorized facial expressions; facial action units performed by professional actors;
occlusions and pose variations; scans with annotated facial landmarks; 2D images available
•
Cons
—medium number of subjects (105), but large number of scans (4666)
•
The Binghamton University 4D Facial Expression (BU-4DFE) Database
To investigate the usability and performance of 3D dynamic facial sequences for facial expres-
sion recognition, a dynamic 3D facial expression database has been created at Binghamton
University (Yin et al., 2008). The Dimensional Imaging's 3D dynamic capturing system
(Di3D, 2006), has been used to capture a sequence of stereo images and produce the depth
map according to a passive stereo-photogrammetry approach. The range maps are then com-
bined to produce a temporally varying sequence of high-resolution 3D images with an RMS
accuracy of 0.2 mm. At the same time, 2D texture videos of the dynamic 3D models are
also recorded. Each participant (subject) was requested to perform the six prototypic expres-
sions (i.e.,
anger
,
disgust
,
fear
,
happiness
,
sadness
, and
surprise
) separately. Each expression
sequence contains neutral expressions in the beginning and the end, so that each expression
was performed gradually from neutral appearance, low intensity, high intensity, and back to
low intensity and neutral. Each 3D sequence captures one expression at a rate of 25 frames
per second and each 3D sequence lasts approximately 4 seconds with about 35,000 vertices
per scan (i.e., 3D
frame
). The database consists of 101 subjects (58 female and 43 male,
with an age range of 18-45 years old) including 606 3D model sequences with 6 prototypic
expressions and a variety of ethnic/racial ancestries (i.e., 28 Asian, 8 African-American, 3
Hispanic/Latino, and 62 Caucasian). More details on the BU-4DFE can be found in (Yin et al.,
2008). An example of a 3D dynamic facial sequence of a subject with “happy” expression
is shown in Figure 5.7, where 2D frames and 3D frames are reported. From left to right, the
frames illustrate the extent of the facial expression from
neutral
to the
onset
,
offset
,
apex
, and
neutral
intensity of the expression.
High Resolution 4D Database from the Applied Digital Signal and Image Processing
Research Centre (Hi4D-ADSIP)
This data set was acquired at the
University of Central Lancashire
and released in late 2011
(Matuszewski et al., 2011; 2012). The dynamic facial acquisition system from the Dimensional
Figure 5.7
Examples of 2D and 3D frames of a 3D video from the BU-4DFE data set
