Facial Animation - Computer Animation: Algorithms and Techniques - page 326

because the model is not captured in the same process as the texture map, registration with a model is an
issue. Using a sequence of images from a video can improve the process.
10.3 Animating the face
Attempts to animate the face raise the questions: What are the primitive motions of the face? And how
many degrees of freedom are there in the face?
10.3.1 Parameterized models
As introduced in the discussion of FACS, parameterizing the facial model according to primitive
actions and then controlling the values of the parameters over time is one of the most common ways
to implement facial animation. Abstractly, any possible or imaginable facial contortion can be consid-
ered as a point in an n -dimensional space of all possible facial poses. Any parameterization of a space
should have complete coverage and be easy to use. Complete coverage means that the space reachable
by (linear) combinations of the parameters includes all (at least most) of the interesting points in that
space. Of course, the definition of the word interesting may vary from application to application, so a
generally useful parameterization includes as much of the space as possible. For a parameterization to
be easy to use, the set of parameters should be as small as possible, the effect of each parameter should
be independent of the effect of any other parameter, and the effect of each parameter should be intu-
itive. Of course, in something as complex as facial animation, attaining all of these objectives is prob-
ably not possible, so determining appropriate trade-offs is an important activity in designing a
parameterization. Animation brings an additional requirement to the table: the animator should be able
to generate common, important, or interesting motions through the space by manipulating one or just a
few parameters.
The most popular parameterized facial model is credited to Parke [ 23 ][ 24 ] [ 25 ] and has already
been discussed in terms of creating facial models based on the so-called conformational parameters
of a generic facial model. In addition to the conformational parameters, there are expression param-
eters . Examples of expression parameters are upper-lip position, eye gaze, jaw rotation, and eyebrow
separation. Figure 10.11 shows a diagram of the parameter set with the (interpolated) expression
parameters identified. Most of the parameters are concerned with the eyes and the mouth, where most
facial expression takes place. With something as complex as the face, it is usually not possible to ani-
mate interesting expressions with a single parameter. Experience with the parameter set is necessary for
understanding the relationship between a parameter and the facial model. Higher level abstractions can
be used to aid in animating common motions.
10.3.2 Blend shapes
The simplest approach to facial animation is to define a set of key poses, also called blend shapes. For a
set of example blend shapes, see Figure 10.12 (Color Plate 10). Facial animation is produced by select-
ing two of the key poses and interpolating between the positions of their corresponding vertices in the
two poses.
This restricts the available motions to be the interpolation from one key pose to another. To gen-
eralize this a bit more, a weighted sum of two or more key poses can be used in which the weights sum