Techniques for Face Motion & Expression Analysis on Monocular Images - 3D Modeling and Animation: Synthesis and Analysis Techniques for the Human Body - page 203

The analysis algorithms presented include those most related to face motion and
expression understanding. Specific image processing can also be used to locate
faces on images, for face recognition intended for biometrics, for general head
tracking and pose deduction, as well as for face animation synthesis. For those
readers acquainted mainly with 3D and graphics, we provide a brief overview of
the most common image processing methods and mathematical tools involved,
pointing to some sources for the algorithmic details that will not be explained or
will be assumed to be known during the description of the state-of-the-art
approaches.
The core of the chapter includes the description of the methods currently being
developed and tested to generate face animation from real face images. The
techniques herein discussed analyze static images and/or video sequences to
obtain general face expressions or explicit face motion parameters. We have
categorized these methods in three groups: “those that retrieve emotion informa-
tion,” “those that obtain parameters related to the Face Animation synthesis
used,” and “those that use explicit face synthesis during image analysis.”
Background
Many video encoders do motion analysis over video sequences to search for
motion information that will help compression. The concept of motion vectors,
first conceived at the time of the development of the first video coding
techniques, is intimately related to motion analysis. These first analysis tech-
niques help to regenerate video sequences as the exact or approximate reproduc-
tion of the original frames by using motion compensation from neighboring
pictures. They are able to compensate for, but not to understand the actions of
the objects moving on the video and, therefore, they cannot restore the object's
movements from a different orientation. Faces play an essential role in human
communication. Consequently, they have been the first objects whose motion
has been studied in order to recreate animation on synthesized models or to
interpret motion for a posteriori use.
Synthetic faces are classified into two major groups: avatars and clones.
Generally, avatars are a rough and symbolic representation of the person, and
their animation is speaker independent because it follows generic rules disre-
garding the individual that they personify. Clones are more realistic and their
animation takes into account the nature of the person and his real movements.
Whether we want to animate avatars or clones, we face a great challenge: the
automatic generation of face animation data. Manually generated animation has
long been used to create completely virtual characters and has also been applied