Automatic 3D Face Model Adaptation with Two Complexity Modes for Visual Communication - 3D Modeling and Animation: Synthesis and Analysis Techniques for the Human Body - page 308

eyebrow is only partly covered by hair, the eyebrow is detected by removing the
covered hair. Otherwise, the eyebrow is completely covered by hair and cannot
be detected. Figure 8 shows the different stages of the eyebrow estimation.
Nose
As nose features, the sides of the nose are estimated (Kampmann & Zhang,
1998). Since the mouth middle position is already determined, search areas for
the sides of the nose are introduced above the mouth. Since the edges of the sides
of the nose have a specific shape, parametric 2D models of the sides of the nose
are introduced. Using this model, each position inside the search area is
evaluated as a possible position of the side of the nose. The positions with the
maximum accumulated value of the luminance gradient at the nose model's
border are chosen as the sides of the nose.
3D Face Model Adaptation
After facial feature estimation is carried out, the two different face models are
adapted in the second step. Using perspective projection, vertices of the 3D face
models which correspond to the estimated 2D facial features are projected onto
the image plane. By comparing these projections with the estimated 2D facial
features in the image, the scaling and facial animation parameters of the face
models are calculated and the face models are adapted.
Low Complexity 3D Face Model Adaptation
For 3D face model adaptation with low complexity, the face model Candide is
exploited. For complexity reasons, the scaling and facial animation parameters
of Candide will be determined with eye and mouth features only (Zhang, 1998).
The scaling parameters for the face model Candide include the scaling
parameters for the face size, for the eye size and for the lip thickness. Before
determining the scaling parameters, the face model Candide is rotated in such
a way that the head tilts of the face model and of the real face in the image match.
The face size is defined by the distance between both eye middle positions, as
well as the distances between the eye and mouth middle positions of the face
model. The 3D eye and mouth middle positions of the face model are projected
onto the image plane. Comparison of the distances of the eyes and the mouth in