Face Detection and Modeling for Recognition - Biometrics Research ...

and nose in frontal views), and because the extraction is time-consuming and usually
generates triangles of the same size in the mesh. Hence, in our model construction, the
desired recognition-oriented facial features can be specified and gradually modified
in the 3D generic face model. The modeling algorithm generates an adapted/learned
3D face model with aligned facial texture. The 2D projections of the texture-mapped
3D model are further used for face verification and recognition.
4.2 Generic Face Model
We choose Waters’ animation model [69], which contains 256 vertices and 441 facets
for one half of the face, because this model captures most of the facial features that are
needed for face recognition (as well as animation), and because triangular meshes are
suitable for free-form surfaces like faces [136]. Figure 4.2 shows the frontal and one side
view of the model, and facial features such as eyes, nose, mouth, face boundary, and
chin. There are openings at both the eyes and the mouth, which can be manipulated.
The Phong-shaded appearance of this model is shown for three different views in
Fig. 4.3.
99