Graphics Reference
In-Depth Information
smaller range, the weights are continuously optimized. Local minima are avoided as described
in Section 4.3.2. The range of the first iteration and the number of new weights tried in each
next iteration were empirically selected as good settings.
4.3.5 Multiple Components
Knowing that the morphable model was generated from 100 3D face scans, an increase of its
expressiveness is most likely necessary to cover a large population. To increase the expres-
siveness, also Blanz and Vetter (1999) proposed to independently fit different components of
the face, namely the eyes, nose, mouth, and the surrounding region. Because each component
is defined by its own linear combination of shape parameters, a larger variety of faces can be
generated with the same model. The fine fitting scheme from the previous section was devel-
oped to be applicable to either the morphable face model as a whole, but also to individual
components of this model.
Algorithm 8 Model fitting (
S
inst
,
scan)
w
range
=
1
.
5,
w
incr
=
0
.
3
for
k
1
to
k
max
do
select vertices (uniform subset of component)
for
i
←
1
to
m
max
do
w
min
=
←
1
w
i
−
w
range
+
2
w
incr
1
w
max
=
w
i
+
w
range
−
2
w
incr
for
w
new
←
w
min
to
w
max
do
morph
S
inst
with
w
new
d
rms
(
S
inst
,
scan) smaller
→
keep
w
new
undo morph
w
new
=
w
incr
morph
S
inst
with
w
i
←
w
new
+
best
w
new
1
w
range
=
w
incr
,
w
incr
=
2
w
incr
return
S
inst
Component selection.
All face instances generated with the morphable model are assumed
to be in correspondence, so a component is simply a subset of vertices in the mean shape
S
(or
any other instance). We define seven components in our adjusted morphable face model (see
Fig. 4.3). Starting with the improved alignment, we can individually fit each of the components
to the scan data using the fine-fitting scheme, obtaining a higher precision of the fitting process
(as shown in Sect. 4.3.6). Individual components for the left and right eyes and cheeks were
selected, so that our method applies to non symmetric faces as well. For comparison, we also
used a face model with four components, one for the eyes, one for the nose, one for the mouth,
and one for the forehead and cheeks. The use of multiple components has no influence on
the fitting time, because the total number of vertices remains the same and only the selected
vertices are modified and evaluated.
