Graphics Reference
In-Depth Information
tion problem. The Active Appearance Model (AAM) [Cootes et al.‚ 1998] and
its variants‚ apply PCA to model both the shape variations of image patches
and their texture variations. They have been shown to be powerful tools for
face alignment‚ recognition‚ and synthesis. Blanz and Vetter [Blanz and Vetter‚
1999] proposed 3D morphable models for 3D faces modeling‚ which model the
variations of both 3D face shape and texture using PCA. The 3D morphable
models have been shown effective in 3D face animation and face recognition
from non-frontal views [Blanz et al.‚ 2002]. Pighin et al. [Pighin et al.‚ 1999]
and Revert et al. [Reveret and Essa‚ 2001] estimated facial deformation based
on the discrepancy between a target face image and the image synthesized from
reference face texture images. Arbitrary facial shapes were approximated by
a linear combination of a set of basic shapes. Furthermore‚ a linear combi-
nation of a set of reference texture images was used to cope with the texture
variations. However‚ the set of reference texture images should be similar to
the target face image‚ for example‚ same person‚ same lighting. Moreover‚
they were computationally expensive because all image pixels are used in the
nonlinear Levenberg-Marquaardt optimization. Recently‚ Liu et al. [Liu et al.‚
2001a] applied both geometric and textural changes to synthesize realistic fa-
cial expressions. The ratio image technique was used to capture the subtle
appearance changes independent of the face surface albedo. In facial expres-
sion classification‚ Tian et al. [Tian et al.‚ 2002] and Zhang et al. [Zhang et al.‚
1998] proposed to train classifiers (e.g. neural networks) using both shape and
texture features. The trained classifiers were shown to outperform classifiers
using shape or texture features only.
In these hybrid approaches‚ some variations of texture are absorbed by shape
variation models. However‚ the potential texture space can still be huge because
many other variations are not modeled by the shape model. Moreover‚ little
has been done to adapt the learned models to new conditions. As a result‚
the application of these methods are limited to conditions similar to those of
training data.
1.3 Issues in flexible appearance model
Because the appearance of facial motions has large variations due to many
factors‚ such poses‚ people and lighting conditions‚ it has been a difficult prob-
lem to adapt appearance models of facials motions to various conditions. In
our framework‚ we focus on (1) the appearance model adaptation for synthesis
over different illumination and people's face albedo; and (2) online appearance
model adaptation during facial motion analysis.
1.3.1 Illumination effects of face appearance
The analysis and synthesis of human faces under arbitrary lighting condi-
tions has been a fascinating yet challenging problem. Despite its difficulty‚
Search WWH ::

Custom Search