Page 35 - IJEEE-2022-Vol18-ISSUE-1

Shakir & Al-Azza
animation streams identified as Charisma, which can be
applied for animation systems utilised with games and virtual
characters in the web.
E. Physics-Based Muscle Modelling
Many attempts have been set on physics-based muscle
modelling to model anatomical facial behaviour. These are
classified into three classes; mass-spring systems, vector
representation, and layered spring meshes. Mass-spring
approaches propagate muscle forces in an elastic spring mesh
which models skin deformation [28]. The vector method
deforms a facial mesh utilising motion fields in delineated
regions of influence [29]. A mass-spring structure was
extended into three connected mesh layers by a layered
spring mesh [30].
The limitation of blendshapes is that they provide only
the linear subspace. Recently, researchers have tended to use
physical simulation to achieve more expressive, non-linear
facial animation. One of the first approaches for physics-
based facial animation was suggested by Sifakis et al. [31],
who construct a detailed face rig comprising of a complete,
anatomically muscle structure, generated manually from the
actor’s medical data. Constructing the muscle structure for
an actor is a time-consuming process. Cong et al. [32]
enhanced an automatic method to transfer anatomy pattern to
target input faces. Ichim et al. [33] fit a template model of
muscles, bones and flesh to face scans. This approach
succeeds by resolving for the muscle activation parameters
that best appropriate the input scans through forward
simulation, and generates an actor physical face mesh for
animation. Ma et al. [34] use a mass-spring system to
construct a blendshape model which incorporates physical
interaction. Kozlov et al. [35] concentrates on the production retargeting. The modelling stage has to do with the model of
of expression-specific physical effects, however the
drawback is that spatially-varying material parameters
require to be painted and set manually for each expression.
Fig. 3: The facial feature points defined in the MPEG-4
standard [36].
Table II
FAP groups in MPEG-4.
Group Number of FAPs
Viseme and expressions 2
Cheeks 4
Eyebrow 8
Tongue 5
Lip, Chin and Jaw 26
| 31
Fig. 4: Example of physics-based facial animation [31].
F. Performance Driven Facial Animation
Performance capture is a method that uses motion
capture technology to represent the performance of the
character. In conventional motion capture, the face and the
body are recorded at different times and then blended
together. Using performance capture, the face and the body
are captured at the same time to describe the entire
performance of the performer. The Polar Express [37] was
the first film to successfully use facial motion capture for an
entire computer graphics (CG) movie as shown in Fig. 5.
Fig. 5: Example of performance-driven facial capture
utilising markers, used in the Polar Express [37].
Artist driven manual key-frame animations may never
capture the subtleties of a human face. The trend in facial
animation has moved towards utilising the human face itself
as the driver and input device for facial animation. Extracting
information from an actual performance of facial movements
is natural, easy, and fast which lead to the concept of
performance-driven facial animation. A ’performance’ can
be understood as a visual capture of an actor’s face talking
and emoting which is utilised to extract information then
retarget the motion onto a digital character. Williams [38]
presented the term performance driven facial animation to
the computer graphics society in Siggraph 1990. Since then
there have been many studies that have extended the main
concept. Hardware motion capture systems were familiar in
the mid-90s and were utilised regularly in short demos [38].
The process of performance driven facial animation can
be divided into three stages: modelling, capture, and
the human face such that it could be digitally stored,
displayed and modified. The choice of representation does
have an effect on the final animation as the model inherently
limits the expressive abilities of the face. modelling
approaches variety from mesh propagation-based
approaches where a single 3D mesh is deformed over the
performance [39, 40] as shown in Fig. 6, 2D and 3D
statistical models based on PCA [41, 42], blendshape models