Introduction To Face Identification

Face identification systems are challenged by variations
in head pose, camera viewpoint, image resolution,
illumination, and facial expression, as well as by
longer-term changes to the hair, skin, and head’s
structure. The geometric approach, which transforms
a face image into simple geometric primitives, holds
the promise of being invariant to illumination and
almost invariant to time-induced changes. Using
well-understood  Multiple View Geometry techniques,
it can also be made practically invariant to
minor pose differences, camera viewpoint changes,
and image resolution. In addition, the geometric approach
has the advantage that a geometric match is
easy to interpret.

Inspite of their intuitive and seemingly precise nature,
geometric face recognition techniques have been
largely replaced by appearance-based techniques. In
these techniques, image representations, which are directly
computed from the pixel-intensities are compared
to estimate similarities between images, or fed
into classifiers that determine the identity of the
person in the image.

Even though the appearance-based techniques are
cleverly designed and engineered, they lack the rigorous
nature of the geometric approach. When an appearance-
based classifier determines a false identify or
wrongly detects a match between two persons, it is
often hard to understand why this happens. Nevertheless,
in 1993 Brunelli and Poggio have shown
that a generic appearance-based method outperforms
a simple geometric-based method on the same dataset,
and contradicting evidence to their finding has
been scarce.