Multi-scale Cartoon Motion Capture and Retargeting without Shape Matching

This paper presents a novel approach for cartoon motion capture and retargeting. Unlike known approaches this method does not rely on finding corresponding points on shape contours. We assume movements have two components, rigid motions and non-rigid deformations. We capture rigid motions by using a similarity transforms. Non-rigid deformations are encoded by the Fourier transform and a key-shape interpolation. An important aspect of the proposed retargeting method is that it is performed in a multi-scale framework which helps us preserve the target´s visual style. We have applied the proposed algorithm to a set of test images and results show very good performance of the proposed method.