The computerized recognition of human faces has long been considered a difficult problem, largely because of the enormous dimension of the "space of distinct faces", or "face space", for short. Now Lawrence Sirovich and Marsha Meytlis of the Mount Sinai School of Medicine in New York City have shown that the dimension of this space is much smaller than had been previously thought.
The key insight is that perceptually, great use is made of the approximate symmetry between the left and right sides of the face. This means that the bulk of the relevant information lies in deviations from perfect symmetry or an artificially symmetrized facial image. This leads not only to a face recognition algorithm that is nearly 100% accurate, but also to a way of generating interesting and realistic synthetic faces.
Further reading
L Sirovich and M Meytlis 2009 PNAS 106 6895.