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

shows the results of local refinement for the face boundary, the nose, the left eye, and
the mouth.
Although our displacement propagation smoothes non-feature skin regions in the
local adaptation, these skin regions can be further updated if a dense range map is
available. However, based on our experiments, we find that the update of non-feature
skin regions does not make a significant difference except in cheek regions because the
displacement propagation already smoothes the skin regions surrounding each facial
feature. Figure 4.10 shows the overlay of the final adapted face model in red and
the target facial measurements in blue. For a comparison with Fig. 4.4, Fig. 4.11
(a)
(b)
Figure 4.10. The adapted model (in red) overlapping the target measurements (in
blue), plotted (a) in 3D; (b) with colored facets at a profile view.
shows the texture-mapped face model. The texture-mapped model is visually similar
to the original face. We further use a face recognition algorithm [78] to demonstrate
the use of 3D model. The training database contains (i) 504 images captured from
28 subjects and (ii) 15 images of one subject generated from our 3D face model,
108