Automatic Cartoon Face Composition Using Caricature Traits

Abstract

We develop a system that can automatically generate a cartoon avatar from a single photograph. This system aims to mimic how artists create a cartoon face with a specific style using a cartoon-face composite application. The cartoon style used in this work was intentionally designed to be very plain and simple in order to bring cuteness to the avatar as well as to add the signature design. This simplicity removes higher details of the face except the prominent features, thus, it causes difficulty in making the cartoon face unique for a human identity. In practice, to make the identity unique for the avatar, artists must match human shapes to their correspondence cartoon shapes from their experience. To mimic this experience, in this work, the mapping operation from human shapes and their cartoon shapes is constructed first. To begin, the training dataset of pairs of a photograph and its artist-created cartoon face is analyzed to capture their patterns. To emphasize more on the facial uniqueness, our “caricature traits” are designed to capture the cartoon style further. Finally, to maintain the cartoon style, our shape search engine will search for the proper cartoon shapes in our pre-fab cartoon libraries. Our evaluation demonstrates that our system can generate a cartoon face for any individual with reasonable similarity comparing to the real artist creation. Cartoon faces from our system are in the form of a 3D model by default. Thus, they can be employed in a variety of applications such as avatars using in games, social network, and virtual reality applications. However, in this work, the main purpose of this application is to 3D-print the model figure of the person.