Graphics Reference
In-Depth Information
To use texture variation to synthesize the mouth interior motion, we sim-
plify the geometry of mouth interior as 3 rows of triangles behind the lips as
in [Guenter et al., 1998]. The original tongue and teeth models are still used
when we do not change the texture of mouth interior.
The ratio image technique assumes a Lambertian manifold, which does not
hold for the interior of mouth. Thus, we choose to directly warp the mouth
interior texture based on inner lip contours to new subject. Because ratio image
technique is applied to transfer the details outside of the inner lip contours,
copying the warped mouth interior texture to the new subject will not create
sharp discrepancy along the inner mouth contour. The mouth interior texture
can be modulated by a constant to account for different camera exposure.
In a preliminary experiment with 2D face images, we map one person's
speech-related appearance variations in mouth area to a different person. The
results are illustrated in Figure 8.11.
Figure 8.11. Mapping visemes of (a) to (b). For (b), the first neutral image is the input, the
other images are synthesized.
3.2 Linear alpha-blending of texture
To synthesize face texture variation caused by facial motion, we use linear
combinations of face appearance examples. This idea of multi-texture blending
has been shown to be useful in facial motion synthesis by Pighin et al. [Pighin
et al., 1998] and Reveret et al. [Reveret and Essa, 2001]. Furthermore, it
is well supported by graphics hardware. For a face appearance example set
we represent an arbitrary texture as
Search WWH ::




Custom Search