Which Ear Regions Contribute to Identification and to Gender Classification?
Di Meng, Sasan Mahmoodi, Mark S. Nixon · 2020
Previous studies in biometrics have shown how gender can be determined from images of ears for recognition, but without specificity. In this paper, we use model-based analysis and deep learning methods for gender classification from ear images. We use these methods to determine the differences between female and male ears. We confirm the identification performance and then the gender discrimination before analyzing which ear parts contribute most to performance. To this end, we compare the heatmaps of different genders with identification heatmaps. It appears from the heatmaps that ears encode females and males differently and we show how this can lead to successful gender discrimination and to increase insight into the process of identification of people by their ears. This could lead to gender identification in surveillance imagery, even when the face is concealed and provides a potential focus for future gender research.