TESTING FACE RECOGNITION SYSTEMS

This paper discusses some of the difficulties which arise when reporti ng computer face recognition testing results. Examples are given, from the authors' own work, of factors that are not usually discussed whic h can have a significant bearing on the reported recognition rate. It is shown that a popular and apparently rigorous approach to testing, w here the choice is between a fixed number of faces, each of which appe ars in each of a fixed number of different conditions, can produce unr epresentative results. The paper raises more general questions, and ma kes recommendations about the type of information that should be inclu ded in order to provide greater confidence in reported recognition rat es.