Modeling and Synthesis of Realistic Visual Speech in 3D - 3D Modeling and Animation: Synthesis and Analysis Techniques for the Human Body - page 286

Figure 15. Independent components yield more intuitive face deformations
than principal components in viseme space. Top: Principal components;
Middle, Bottom: Independent components.
the variation. The advantage, rather, is the more intuitive deformations that
correspond to the independent components, where each stays closer to a single,
anatomical action of the face.
Finally, on a more informal score, we found that only about one or two PCs could
be easily described, e.g., “opening the mouth.” In the case of ICs, six or so
components could be described in simple terms. Figure 15 shows a comparison
between principal and independent components. In both cases, there is a
component that one could describe as opening the mouth. When it comes to a
simple action, like rounding the mouth, there is a single IC that corresponds to this
effect. But, in the case of PCs, this rounding is never found in isolation, but is
combined with the opening of the mouth or other effects. Similar observations
can be made for the other ICs and PCs.
One could argue that animation can proceed directly and uniquely as a combi-
nation of basic modes (e.g., independent components) and that going via visemes
is an unnecessary detour. Discussions with animators made it clear, however,
that they insist on having intuitive keyframes, like visemes and basic emotions,
as the primary interface. Hence, we give animators control both at the level of