Artistic Rendering of Portraits

Abstract

This chapter presents an overview of the latest artistic portrait rendering techniques. Artistic rendering or portraits can be viewed as an image generating process controlled by two factors. The first one is the face fidelity, namely, a rendered portrait image should preserve a certain amount of the original face's information. The second factor is the artistic style, for example, sketch, painting, etc. In the literature, different portrait rendering algorithms either adopt different models and data structures to represent the facial information, or use different graphical elements and compositional methods to simulate various styles. These two factors essentially reveal the two principles in portraiture, namely the pursuit of likeness and aesthetic.

NPRportrait 1.0: A three-level benchmark for non-photorealistic rendering of portraits

“…Kyprianidis et al's [1] assessment from 2013 still holds true: "Portraits are an example of subject matter currently rendered poorly by general purpose stylization algorithms since they are particularly sensitive to distortion or detail loss in facial regions." We refer the reader to Zhao and Zhu's work [36] for an overview of portrait-specific NPR methods prior to deep learning, and to Yaniv et al's paper [35] for references to more recent methods. In this section we very briefly outline the 11 NPR algorithms which are evaluated in Section 4.…”