Deep Facial Features for Analysing Artistic Depictions - A Case Study in Evaluating 16th and 17th Century Old Master Portraits

Abstract

Convolutional neural network (CNN) based deep learning has recently become the standard de facto for computer assisted image analysis and classification. In this work, we use an in-house trained deep face recognition model to extract facial features from images of old portraits to compare their degree of similarity. Taking the well-known Visual Geometry Group (VGG) deep learning model as the basis, our in-house trained model is fine-tuned for enhanced facial similarity analysis, providing particular attention to the effects from prominent parts of the face, such as the eyes, nose and mouth features. We show how this model can be efficiently utilised to evaluate faces present in age-old master portraits. More specifically, we undertake facial similarity analysis of the faces in the oil paintings of Madonna and Child of de Brécy Tondo and Sistine Madonna by Raphael, of which the former has been the subject of national and international research for nearly 40 years.