Generalizable deep features for ocular biometrics

摘要

There has been a continued interest in learning features that are generalizable across sensors and spectra for ocular biometrics. This is usually facilitated through a model that can learn features that are robust across pose, lighting conditions, spectra, and device sensor variations. In this paper, we propose an efficient deep learning-based feature extraction pipeline for learning the aforementioned generalizable features for ocular recognition. The proposed pipeline uses a relatively small Convolutional Neural Network (CNN) based feature extraction model along with a region of interest (ROI) detector and data augmenter. Our proposed CNN model has 36 times fewer parameters compared to the popular ResNet-50. Cross dataset experiments on five benchmark datasets suggest that the proposed feature extraction model, trained only on 200 subjects from the VISOB dataset, reduces the error rate up to 7 × when compared to the existing models.