Graph based transductive learning for cartoon correspondence construction

Correspondence construction of characters in key frames is the prerequisite for cartoon animations' automatic inbetweening and coloring. Since each frame of an animation consists of multiple layers, characters are complicated in terms of shape and structure. Therefore, existing shape matching algorithms, specifically designed for simple structures such as a single closed contour, cannot perform well on characters constructed by multiple contours. This paper proposes an automatic cartoon correspondence construction approach with iterative graph based transductive learning (Graph-TL) and distance metric learning (DML) estimation. In details, this new method defines correspondence construction as a many-to-many labeling problem, which assigns the points from one key frame into the points from another key frame. Then, to refine the correspondence construction, we adopt an iterative optimization scheme to alternatively carry out the Graph-TL and DML estimation. In addition, in this paper, we adopt the local shape descriptor for cartoon application, which can successfully achieve rotation and scale invariance in cartoon matching. Plenty of experimental results on our cartoon dataset, which is built upon industrial production suggest the effectiveness of the proposed methods for constructing correspondences of complicated characters.