Data augmentation in CNN-based periocular authentication
Ryan Dellana, Kaushik Roy · 2016
In this research, we apply a Convolutional Neural Network (CNN) to periocular authentication on two datasets. To increase accuracy, we try a variety of data augmentation techniques and compare their relative benefits. We find that, with augmentation appropriate to the dataset, CNN accuracy may be comparable to or significantly higher than traditional methods like Local Binary Pattern Histograms (LBPH) and Eigenface.