Facial Animation - Computer Animation: Algorithms and Techniques - page 320

FIGURE 10.5
Simple facial model using rigid components for animation.
(Image courtesy of John Parent.)
The first problem confronting an animator in facial animation is creating the geometry of the facial
model to make it suitable for animation. This in itself can be very difficult. Facial animation models
vary widely from simple geometry to anatomy based. Generally, the complexity is dictated by the
intended use. When deciding on the construction of the model, important factors are geometry data
acquisition method, motion control and corresponding data acquisition method, rendering quality,
and motion quality. The first factor concerns the method by which the actual geometry of the head
of the subject or character is obtained. The second factor concerns the method by which the data
describing changes to the geometry are obtained. The quality of the rendered image with respect to
smoothness and surface attributes is the third concern. The final concern is the corresponding quality
of the computed motion.
The model can be discussed in terms of its static properties and its dynamic properties. The statics
deal with the geometry of the model in its neutral form, while the dynamics deal with the deformation
of the geometry of the model during animation. Facial models are either polygonal or higher order.
Polygonal models are used most often for their simplicity (e.g., [ 24 ] [ 25 ] [ 32 ][ 33 ]); splines are chosen
when a smooth surface is desired.
Polygonal models are relatively easy to create and can be deformed easily. However, the smooth-
ness of the surface is directly related to the complexity of the model, and polygonal models are visually
inferior to other methods of modeling the facial surface. Currently, data acquisition methods only sam-
ple the surface, producing discrete data, and surface fitting techniques are subsequently applied.
Spline models typically use bicubic, quadrilateral surface patches, such as Bezier or B-spline, to rep-
resent the face.While surface patches offer the advantage of lowdata complexity in comparison to polyg-
onal techniques when generating smooth surfaces, they have several disadvantages when it comes to
modeling an object such as the human face. With standard surface patch technology, a rectangular grid
of control points is used to model the entire object. As a result it is difficult to maintain low data com-
plexity while incorporating small details and sharp localized features, because entire rows and/or entire
columns of control information must be added. Thus, a small addition to one local area of the surface to
better represent a facial feature means that information has to be added across the entire surface.
Hierarchical B-splines, introduced by Forsey and Bartels [ 16 ] , are a mechanism by which local
detail can be added to a B-spline surface while avoiding the global modifications required by standard
B-splines. Finer resolution control points are carefully laid over the coarser surface while continuity is
carefully maintained. In this way, local detail can be added to a surface. The organization is hierarchi-
cal, so finer and finer detail can be added. The detail is defined relative to the coarser surface so that
editing can take place at any level.