An approach to labeling the components of human faces from range image
s is proposed. The components of interest are those humans usually fin
d significant for recognition. To cope with the nonrigidity of faces,
an entirely qualitative approach is used. The preprocessing stage empl
oys a multistage diffusion process to identify convexity and concavity
points. These points are grouped into components and qualitative reas
oning about possible interpretations of the components is performed. C
onsistency of hypothesized interpretations is carried out using contex
t-based reasoning. Experimental results on real range images of severa
l faces are provided. (C) 1994 Academic Press, Inc.