Gender and age group classification from multiple soft biometrics traits

We compare the classification accuracies of estimating the global human demographic attributes - gender and age group from three gender and age models trained with hand, voice recordings and fingerprint biometric characteristics, respectively. Biometric data was acquired from the same subjects within six months. Training and test sets were extracted from the acquired datasets. We show that classification accuracy can be improved by fusing scores of the predictions from the three gender models as well as the three age models at the score level using the sum rule. The models were evaluated with disjointed test sets. The results of predicting gender from the three biometric characteristics show a ranking of classification performance in this order: hand, voice and fingerprint. We also observe that fusing the classifier models improves and consolidates classification accuracy. Finally, we propose three new datasets of hand, voice and fingerprint biometrics, different from existing datasets.