AniME (stress on the "me") is a Python pipeline which, given a user input photo, automatically generates a corresponding stylized anime portrait by template matching to a hand-drawn (by me!) library. I worked on it with my friend Michelle Nguyen as our final project for our computational photography course. We wanted to combine vision algorithms to generate a fun, speculative system paying homage to the media we grew up watching. Our 4-page paper is available here, and our code is here.


Abstract

The popularity of anime (Japanese animated productions) has transcended cultural boundaries, resulting in high volumes of fan works such as personalized avatars. However, drawing aesthetically pleasing anime is a craft that takes a long time to develop. AniME (pronounced ani-mee) is a pipeline which integrates computer vision, computational photography, and data-driven frameworks to automatically generate an anime-style portrait from a user given photo. Our approach renders the corresponding lineart using landmark facial annotations, while features such as eye shape and hairstyle are obtained via template matching to a hand-drawn library of anime eyes and hairstyles. Finally, colors of the anime portraiture are determined through simple averaging and clustering of selected image patches. We validate our tool using photos of people of varying genders, ethnicity, and hair styles.

Credits
Library images, accessories (eyes/eyebrows/glasses/ears) algorithms: Jingyi Li
Hair and face shape algorithms: Michelle Nguyen