Figure 8.10. Recognition error rate comparison of before relighting and after relighting on the
Yale face database.
the input image by using techniques such as those reported in [Blanz and Vetter,
1999, Zhang et al., 2001].
3. Synthesize Appearance Details of Facial Motion
Liu et al. [Liu et al., 2001a] used ratio image technique, called Expression
Ratio Image (ERI), to map one person's face expression details to other people's
faces. More generally, this technique can be used to synthesize facial-motion-
related appearance variations for new subject. In this way, the appearance
examples for visemes and expressions can be used in face synthesis for new
people. In this section, we discuss two issues that need extra consideration when
applying the ratio image technique. First we discuss how to map appearance
of mouth interior in Section 3.1. The mouth interior appearance is important
for applications such as lip-reading, but does not satisfy the assumption of
ratio image technique. Next, We describe how to create face animations with
appearance variations in Section 3.2.
3.1 Appearance of mouth interior
The appearance of teeth and tongue is important for visual speech perception.
However, there is no robust technique to capture their motion so far. Moreover,
compared to the skin it is more difficult to measure the surface reflectance
property for realistic rendering. Due to these reasons, it is more feasible to
capture the motion details of mouth interior using appearance.