The Role of Facial Hair on Roman Emperors' Face Recognition

Hairstyle plays an important role in the identification of Roman imperial portraits in the historical and archeological context. This makes sense because, at the time, hairstyle and facial hair were considered important in projecting an image of authority and legitimacy, and this image is expected to be applied consistently. However, deep learning face recognition models have shown that said portraits can also be recognized without explicitly focusing on the hairstyle. In this paper, we demonstrate that such models still achieve reasonable recognition rates if the coiffure and facial hair of the Roman emperor portraits are removed from the input image by replacing the relevant areas with average face image data, generated from the Roman emperor's dataset. The comparison of the recognition performance between the datasets with and without coiffure and facial hair shows that these features are not essential for reasonable recognition rates for a face recognition system.