Abstract
The human face is one of the most ecologically relevant objects for visual perception.
Although the face changes expressions constantly and in a variety of complex ways, we
are able to interpret these with a quick glance at a face. In particular, facial motion
plays a complex and important role in communication. It can be used, for example, to
convey meaning, to express an emotion or to modify the meaning of what is said. My
research is focused on what we can learn, using psychophysical methodologies, about
the human visual system from the way faces move. I will attempt to develop a detailed
cognitive model for the perception of expressions by exploring and differentiating the
information channels contained in facial expressions. Here, I present the results of
psychophysical experiments, in which we manipulated real video sequences of facial
expressions of different actors. In the first experiment, we scaled down the video
sequences to find out how the recognition of an expression depends on the presented
image size [2]. In a second set of experiments, Cunningham et al. selectively ’froze’
portions of a face to produce an initial, systematic description of the parts of a face that
are necessary and sufficient for the recognition of facial expressions [3]. Based on these
experiments, I will outline future work in which we plan to use computer animated
faces [1]. This will allow us to produce realistic image sequences while retaining
complete control over what occurs in the images (e.g., to finely alter the temporal
parameters such as the speed, acceleration, duration, or synchronization of facial
motion). Finally, I want to propose a unifying framework of interpretation and
manipulation of facial analysis and synthesis, which contains different, hierarchically
organized levels of perception and simulation. Within this framework, we can
systematically identify and analyze the information channels that are addressed by the
cognitive experiments described above. The results from this line of research are
expected not only to shed light on perceptual mechanisms of expression recognition,
but also to help improve computer animation in order to create perceptually consistent,
realistic and believable conversational agents.
