Game Development Reference
In-Depth Information
the same, but they should be different to the mean of chrominance in the mouth-
inside. The variances are utilized to describe the uniformity of the chrominance
values
U
within these areas. The term
f
(
d
,
d
,
o
,
o
)
consists of the
open
2
u
o
u
o
g
Y
along lip contours
W
i
. The
run and length of the parabolic curves are defined with a 2D
mouth-open
model
and are dependent on the parameters to be estimated.
The parameters
(
d
u
, d
o
, o
u
, o
o
)
in the
mouth-open
model are determined by
minimization of the cost
(
x
,
y
)
addends of edge strength (image gradient)
(
,
,
,
)
df
. To reduce the computational
complexity, the cost function is only evaluated at the already detected candidates
for the lip contours (ref. Figure 6). From all possible combinations of the lip
contour's candidates, the combination with the least cost is determined as the
estimates for the lip thickness and the lip opening heights in the
mouth-open
model.
d
o
o
open
u
o
u
o
Other Facial Features
In case of the high complexity mode, other facial features besides the eyes and
the mouth are estimated.
Chin and cheek contours
For the estimation of chin and cheek contours, the approach described in
Kampmann (2002) is used. The chin and the cheeks are represented by a
parametric 2D model. This parametric model consists of four parabola branches
linked together. The two lower parabola branches represent the chin, the two
upper parabola branches define the left and the right cheek. Taking into account
the estimated eye and mouth middle positions, search areas for the chin and
cheek contours are established. Inside each search area, the probability of the
Figure 7. Estimation of chin and cheek contours: (a) Estimated chin
contour, (b) Estimated chin and cheek contours.
(a) (b)
Search WWH ::
Custom Search