Face-based gender and ethnicity classification using golden ratio template

Abstract

There has been increasing interest of video surveillance systems, especially in places in need of high security, to reduce the crime rate. The most informative biometric modality for human identification is human face. The face carries several types of information of a person besides the identity such as his gender, age, ethnicity and emotion. Visual information such as gender, age and ethnicity, in other words soft biometric modals, play critical roles in human identification as well as face to face communication. Most gender and ethnicity recognition methods use the full face data assuming each face part has equal discriminant capability identification and recognition. In this research, we improve the gender and ethnicity recognition by employing the optimum decision making rule based on the confidence of 16 different face regions. Automatic facial patch extraction is done by the modified golden ratio template on the full face aligned with multiple base points to prevent the displacement of facial parts due to different facial part distances of the people in different genders and ethnicities.