Demographic classification: Do gender and ethnicity affect each other?

Gender and ethnicity classification are challenging topics in the field of face analysis. Some features, like skin color, are relevant only for ethnicity but not for gender; some others, like face geometry, are important for both. The impact of ethnicity in gender perception, as the effect of gender on ethnicity disambiguation, is not clear. This paper provides a study to check if gender and ethnicity affect each other during the classification. Three different well-established algorithms have been implemented to provide significant experiments. These algorithms are used for both gender and ethnicity classification. Ethnicity-specific gender classifiers are trained and tested using faces from a specific ethnicity; the accuracies achieved are compared with the ones obtained using generic gender classifiers, trained and tested with faces from different ethnic groups. With a similar procedure we compare gender-specific ethnicity classifiers, trained and tested selecting faces with a specific gender, with generic ethnicity classifiers, trained and tested with both male and female faces. The study shows that specific and generic classifiers perform equally. That means, for the features selected, gender and ethnicity do not affect each other.