Graphics Reference
In-Depth Information
Figure 4.1
Face segmentation. The depth image (left) is converted to a surface mesh (middle). The
surface mesh is cleaned, the tip of the nose is detected and the face segmented (right, in pink)
the pose of the face and localizes the tip of the nose. Before pose normalization, we applied
a few basic preprocessing steps to the scan data: The 2D depth images were converted to
triangle meshes by connecting the adjacent depth samples with triangles, slender triangles and
singularities were removed, and only considerably large connected components were retained.
Afterwards, the face is segmented by removing the scan data with a Euclidean distance larger
than 110 mm from the nose tip. The face segmentation is visualized in Figure 4.1.
4.3 Face Model Fitting
In general, 3D range scans suffer from noise, outliers, and missing data, and their resolution
may vary. The problem with single face scans, the GAVAB scans in particular, is that large
areas of the face are missing, which are hard to fill using simple hole filling techniques. When
the morphable face model is fitted to a 3D face scan, a model is obtained that has no holes, has
a proper topology, and has an assured resolution. By adjusting the
m
99 weights
w
i
for the
eigenvectors, the morphable model creates a new face instance. To fit the morphable model to
3D scan data, we need to find the optimal set of
m
weights
w
i
. This section describes a fully
automatic method that efficiently finds a proper model of the face scan in the
m
-dimensional
space.
=
4.3.1 Distance Measure
To evaluate if an instance of the morphable face model is a good approximation of the 3D face
scan, we use the root mean square (RMS) distance of the instance's vertices to their closest
points in the face scan. For each vertex point (
p
) from the instance (
M
1
), we find the vertex
point (
p
) in the scan data (
M
2
) with the minimal Euclidean distance
p
)
e
min
(
p
,
M
2
)
=
min
p
∈
M
2
d
(
p
,
(4.1)
using a kD-tree. The RMS distance is then measured between
M
1
and
M
2
as:
n
1
n
d
rms
(
M
1
,
M
2
)
=
e
min
(
p
i
,
M
2
)
2
(4.2)
i
=
1
