Face Verification in Practice: The Case of Greek Artist Leonidas Arniotis

Abstract

In this paper we investigate a specific case of face verification: Given some confirmed photos and video frames of Leonidas Arniotis, a Greek artist of late 1800 and early 1900s, we have examined two photos, of very low resolution, of an unknown person that could be Leonidas Arniotis. The widespread of deep learning and the availability of ResNet-50 model, which was trained on VGGFace2 dataset, allowed us to apply transfer learning for face verification purposes. At the same time an alternative technique which is based on facial points matching was applied. The main aim of the paper is to show that despite the progress made in the fields of face recognition and face verification, actual systems that can be used by non-experts for examining real world cases, like the one we deal in the current paper, remain impractical. In the framework of the current work, we also examine and discuss, in a critical way, the effectiveness of transfer learning for real face verification cases through specific examples.