Pattern Recognition Letters ( IF 3.255 ) Pub Date : 2020-10-17 , DOI: 10.1016/j.patrec.2020.10.009 Lucia Cascone; Carlo Medaglia; Michele Nappi; Fabio Narducci
The pupil size is a soft biometric trait but in-depth study to analyze it for biometric purposes is lacking in the literature, as well as datasets focused on this field of research are missing. On the basis of these observations, we present an extensive study with the objective of demonstrating that pupil size and dilation over time can be potentially used to classify people by age and by gender. To do this, 14 supervised classifiers were applied on a dataset meant for gaze analysis. Measuring the right and left pupil individually and also simultaneously, the performances of the classifiers have been compared and the worst and best performing selected to support potential fusion strategies. If good results have been obtained for age classification, ranging in 79% - 82% of accuracy, it cannot be said the same for gender. This is due to the dual nature of this biometric trait. Pupil size can be considered as a physical trait for the age classification and behavioral trait for the gender. The results achieved in this study suggest the need for more balanced and reliable datasets.