Page 33 - IJEEE-2022-Vol18-ISSUE-1
P. 33
Shakir & Al-Azza | 29
models, muscle-based. Blendshapes is a simple linear model Fig. 2: Linear interpolation is performed on blendshapes.
of facial expression. It has driven animated characters in
Hollywood movies, and is a standard feature of commercial B. Parameterisation
animation packages. The beginning of the blendshape The essential concept for a parameterised model is to
method is in the computer graphics industry by the 1980s. In
1972, Frederick I. Parke first utilised the shape interpolations build a desired face or facial expression relied on some
and animated the face utilising cosine interpolation. By the number of controlling parameter values. If the model
late 1980s the offset blendshape scheme became widespread supports conformation parameters, it can be utilised to
and exist in commercial software [1]. In this variant a neutral generate a range of individual faces.
facial shape is designated and the residual shapes are
substituted by the differences between those shapes and the Parameterisations identify any face and expression by a
neutral one. The differences between the target shape and the grouping of independent parameter values [7]. Different than
neutral face are restricted to a small region, although it relies interpolation approaches, parameterisations permit explicit
on the modeller to generate shapes with this property. This control of particular facial configurations. Groupings of
concept was expanded to a segmented face where separate parameters offer a large range of facial expressions with low
parts are blended independently [2], therefore promising computational costs. Frederic I. Parke produced the first 3D
local control. A standard instance is the segmentation of a parametric model of a human face [8]. The facial geometry
face into an upper part and a lower part: the upper part is is broken into parts and controlled by parameters, for
utilised for expressing emotions, whereas the lower part instance, the rotation of the jaw or the direction of an eye’s
expresses speech [3]. Despite the blendshape method is gaze. The parameters affecting the various nodes are showed.
conceptually simple, developing such a blendshape face A parametric facial animation system describes a set of
model is a labor intensive effort. In order to express a whole parameters for the face. These are mostly the expression
collection of realistic expressions, digital modellers often parameters for different fragments of the face, for example
have to produce large libraries of blendshape targets. For forehead, eyes, and nose and the conformation parameters
instance, the character of Gollum in The Lord of the Rings that globally applied to the entire face. The main parameters
had 675 targets [4]. Generating a realistically detailed model for the forehead are scale and shape of forehead. The main
can be as much as a year of effort for a skilled modeller, parameters for the eyes are pupil size, eyelid opening,
including many iterations of enhancement. The Blendshape eyebrow colour and separation, etc. The conformation
facial animation is the general choice for realistic human parameters are aspect ratio of the face, colour of the skin, etc.
characters in the films production. The method has been Each expression parameter affects a set of vertices of the face
utilised for lead characters in films such as King Kong [5], model. In this way, key frames can be defined easily. A
The Curious Case of Benjamin Button [6]. The simplest case complete generic parameterisation is not possible, because
is an interpolation between two key-frames at extreme that the parameter set relies on the facial mesh topology.
positions over a time interval as shown in Fig. 2. Moreover, boring manual tuning is required to set parameter
values. The limitations of parameterisation led to the
Fig. 1: Overview flows of facial animation. development of different approaches such as morphing
between images and geometry, physically muscle-based
animation, and performance driven animation.
C. Facial Action Coding System
The most common expression coding systems are facial
action coding system (FACS) and moving pictures experts’
group-4 (MPEG-4) Facial animation models. FACS was
suggested by Ekman and Friesen in 1978 [9] and has been
improved in 2002 [10]. It defines all the movements that can
be seen in the face based on face anatomy. It has been utilised
extensively in facial animation during the last few decades.
FACS has become a typical in interpretation facial behaviour
in science research such as psychophysiology [11], and in
fields such as video games [12], movies [6], robots [13] and
facial expression recognition, mapping, generation, [14–16].
FACS is an anatomically based system for defining all
observable face movements. Each component of a facial
