Automatic 3D Face Model Adaptation with Two Complexity Modes for Visual Communication - 3D Modeling and Animation: Synthesis and Analysis Techniques for the Human Body - page 309

the image with those of the projections of Candide yields the scaling factors for
the face size. The eye size is defined as the distance between both eye corner
positions. The scaling parameter for the eye size is determined by comparing the
projections of the 3D eye corner positions of the face model onto the image plane
with the estimated 2D eye corner positions. As the scaling parameter for the lip
thickness, displaced vectors of the lip's vertices of the face model are intro-
duced. The vertices that represent the inside contours of the lips are fixed during
the scaling of the lip thickness. The vertices that describe the outside contours
of the lips are shifted outwards for the scaling of the lip thickness. The eye
opening is defined by the positions of the upper and lower eyelids. The position
of the upper eyelid can be changed by AU 41 and that of the lower eyelid by AU 7 .
Since the 3D face model has fully opened eyes at the beginning, the eyelids of
the face model are closed down, so that the opening heights of the face model
match the estimated eyelid opening heights in the image plane. The values of
AU 41 and AU 7 are calculated by determining the range of the movement of the
eyelids of the face model. The movement of the mouth corners are represented
by AU 8 and AU 12 . AU 8 moves the mouth corners inward and AU 12 moves them
outward. For the determination of these two Action Units' values, the estimated
mouth corner positions are utilized. The mouth opening is defined by the position
and movement of the upper and lower lips. The position of the upper lip is
determined by AU 10 and that of the lower lip by AU 16 . For the determination of
these two Action Units' values, the estimated 2D opening heights of the lips are
used.
High Complexity 3D Face Model Adaptation
For adaptation of the high complexity face model, all estimated facial features
(eyes, mouth, eyebrows, nose, chin and cheek contours) are taken into account.
Using all estimated facial features, scaling and initial facial animation parameters
of the high complexity face model are calculated and used for the face model
adaptation to the individual face. First, orientation, size and position of the face
model are adapted (only eye and mouth middle positions and cheek contours are
used here). Then, the jaw of the face model is rotated. In the next step, the chin
and cheek contour of face model is adapted. Finally, scaling and facial animation
parameters for the rest of the facial features (eyes, mouth, eyebrows, nose) are
determined.
For orientation, the lateral head rotation is adapted first. The quotient of the
distances between eye middle positions and cheek contours on both sides of the
face is introduced as a measure for the lateral head rotation. For adaptation, the
face model is rotated around its vertical axis as long as this quotient measured
in the image plane using the estimated facial features is the same as the quotient