Understanding critical factors in gender recognition

Gender classification is a task of paramount importance in face recognition research, and it is potentially useful in a large set of applications. In this paper we investigate the gender classification problem by an extended empirical analysis on the Face Recognition Grand Challenge version 2.0 dataset (FRGC2.0). We propose challenging experimental protocols over the dimensions of FRGC2.0 – i.e., subject, face expression, race, controlled or uncontrolled environment. We evaluate our protocols with respect to several classification algorithms, and processing different types of features, like Gabor and LBP. Our results show that gender classification is independent from factors like the race of the subject, face expressions, and variations of controlled illumination conditions. We also report that Gabor features seem to be more robust than LBPs in the case of uncontrolled environment.