The local search for maximum gray-level gradients is guided by the globally
learned lip shape space.
Yullie et al. [Yullie et al., 1992] define a facial feature as a deformable tem-
plate, which includes a parametric geometrical model and an imaging model.
Deformable template poses tracking as an analysis-by-synthesis problem. The
geometrical model describes how the shape of the template can be deformed
and is used to measure shape distance from the template. The imaging model
describes how to generate an instance of the template and is used to measure
the intensity distance from the template. An energy function is designed to link
different types of low-level image features, such as intensity, peaks, and edges,
to the corresponding properties of the template. The parameters of the template
are calculated by steepest descent. Nonetheless, the parametric facial feature
models are usually defined subjectively.
3D parameterized model
DeCarlo and Mataxas [DeCarlo and Metaxas, 2000] propose an approach
that combines a deformable model space and multiple image cues (optical flow
and edge information) to track facial motions. The edge information used is
chosen around certain facial features, such as the boundary of the lips and eyes.
To avoid high computation complexity, optical flow is calculated only for a set
of image pixels. Those image pixels are chosen in the region covered by the face
model using the method proposed by Shi and Tomasi [Shi and Tomasi, 1994].
The deformable model is a parametric geometric mesh model. The parameters
are manually designed based on a set of anthropometric measurements of the
face. By changing the values of the parameters, the user can obtain a differ-
ent basic face shape and deform the basic face shape locally. The deformable
model in [DeCarlo and Metaxas, 2000] helps to prevent producing unlikely
facial shapes during tracking. However, it is labor-intensive to construct the
face deformation model, and some facial deformation (e.g. lip deformations
produced during speech) may not be represented adequately using the anthro-
A number of researchers have proposed to model facial deformation using
FACS system. The FACS based 3D models impose constraints on the sub-
space of the plausible facial shapes. The motion parameters include global
face motion parameters (rotation and translation) and local facial deformation
parameters, which correspond to the weights of AUs in [Li et al., 1993, Tao
and Huang, 1999] and to the FACS-like control parameters in [Essa and Pent-