Correspondence specification learned from master frames for automatic inbetweening

Bingwen Jin, Weidong Geng

Abstract: We present a new approach to automatically specify the correspondences between hand-drawn keyframes for automatic inbetweening in 2D facial cartoon animation. Recent techniques are not robust to occlusion as they mainly rely on the information provided by the keyframes, which are 2D drawings lacking 3D information. Our approach creates a hybrid face model to learn the prior knowledge about the approximate 3D geometry and multi-view appearances of an individual character’s face from master frames to overcome the lack of information. Based on the hybrid model, we combine our example-based stroke annotating method with our example-based viewpoint recognition method to automatically specify accurate stroke correspondences, even when there is occlusion. Our approach facilitates auto-inbetweening much.