Graphics Reference
In-Depth Information
instance were distorted. The post processing method to blend the borders of the components
introduces a new set of surface samples without correspondence to the model either.
We present a new bootstrapping algorithm that can be applied to general 3D face data. Our
algorithm automatically detects if a new face scan cannot be sufficiently modeled, establishes
full correspondence between the face scan and the model, and enhances the model with this
new face data. Without the use of unreliable optic flow (Basso et al., 2006) or semi-automatic
non rigid ICP (Amberg et al., 2007), we are able to bootstrap the 3Dmorphable face model with
highly accurate face instances. As a proof of concept, we (1) fit the initial morphable face model
to several 3D face scans using multiple components, (2) blend the components at the borders
such that accurate point-to-point correspondences with the model are established, (3) add the
fitted face instances to the morphable model, and (4) fit the enhanced morphable model to the
scan data as one single component. In the end, we compare each single component fit obtained
with the enhanced morphable model to the single component fit obtained with the initial
morphable model. Qualitative and quantitative evaluation shows that the new face instances
have accurate point-to-point correspondences that can be added to the initial morphable face
model. By comparing the multiple and single component fit, our bootstrapping algorithm
automatically distinguishes between new face data to add and redundant data to reject. This
is important to keep both the model fitting and the face recognition with model coefficients
time-efficient.
4.4.1 Bootstrapping Algorithm
We fit the morphable face model to 3D scan data to acquire full correspondence between
the scan and the model. We crop the morphable face model and lower its resolution so
that
n
=
12
,
964 vertices remain for the fitting. We use the new set of correspondences
S
=
z
n
)
T
3
n
(
x
1
,
y
1
,
z
1
,...,
x
n
,
y
n
,
∈
to automatically bootstrap the model, in order to increase
its expressiveness.
Figure 4.5 shows the changes of the original face model when the weight of the third
eigenvector (
w
3
) is varied. It can be noticed that the model is tilted upwards and downwards.
This variation in one of the first eigenvectors means that the alignment of the 100 sets
S
i
is
not optimal for face identification using model coefficients. To adjust the original model, we
realigned each reduced face shape
S
i
to the mean face
S
of the morphable model using the
ICP algorithm, and recomputed the PCA model. Visual inspection of our newly constructed
PCA model showed no signs of pose variations.
The main problem in bootstrapping the 3Dmorphable face model, is that (1) we only want to
add example faces that are not covered by the current model, (2) new example faces suffer from
Figure 4.5
Changingweight
w
3
{
−
2,0,
+
2
}
causes an unwanted change in the gaze direction. Copyright
C
2009, IEEE
