Convolutional neural network-based periocular recognition in surveillance environments

Visible light surveillance cameras are currently deployed on a large scale to prevent crime and accidents in public urban environments. For this reason, various human identification studies using biometric data are underway in surveillance environments. The most active research area is face recognition, which generally shows excellent performance; however, aging, changes in facial expression, and occlusions by accessories cause a rapid decline in recognition performance. To resolve these problems, we propose a periocular recognition method in surveillance environments that is based on the convolutional neural network. In this paper, experiments were performed using the custom-made Dongguk periocular database and the open database of ChokePoint database. It was confirmed that the proposed method performs better than existing techniques used in periocular recognition. It was also found to perform better than conventional techniques in face recognition when an occlusion is present.