Entropy in Biometric Face Template Analysis

We will consider systems whose galleries contain more templates for each subject, and will explore the concept of representativeness of a biometric sample. This parameter varies from subject to subject, and is not a feature of the whole gallery. The gallery samples of certain subject might be too much similar, just because they are all of excellent quality, due to a systematic acquisitions in well-controlled conditions. In this situation, even a moderately different sample probe of the subject in input will cause an error, even by an accurate system, if the excellent conditions are not maintained. This seems to indicate that quality measures alone cannot guarantee good performances. Correct recognition in different situations may require a sufficient amount of variation in gallery templates. We call this (subject’s) gallery feature representativeness, and investigate the role of mutual information/entropy in defining it.