Exaggeration of Facial Features in Caricaturing

Abstract

A new NPR system is proposed in this paper to generate a facial caricature from a 2–D picture. In human perception, we recognize a person by seeing his feature parts and the outline of his face. Our system can exaggerate the feature parts and still keep the overall impression. First, we must have the information about the outlines of the facial features to determine the characteristics of the features. Because the facial features are too complicated, we obtain the information of the features manually. We can simulate the actual outlines of facial features by knowing only some key points. After obtaining those key points of facial features, we can decide which facial feature should be exaggerated based on rules. Finally our system renders a new pictorial in a non–photorealistic rendering (NPR) system using bold edges and large regions of constant shade.