PDF version of this entire document
I n order to study the accuracy of 3-D face recognition algorithms, one must differentiate between the facial expression as a contributor to variation and the physical component which hardly ever varies. The former can help one learn something about a person at one given point in time, whereas the latter helps distinguish between people. In order to remove variational impact caused primarily by facial expressions, one can assume a commonality between facial expressions and study their statistical nature automatically. If faces have expression-free (or neutral) equivalents, it then becomes abundantly clear how to tell faces apart. A good solution in such a problem-inducing situation would be a framework that can separate the contribution of expression from the contribution that hardly ever varies. Then, the framework also becomes more able and better equipped to annul the former component.
