Globally Optimal Toon Tracking

摘要

The ability to identify objects or region correspondences betweenconsecutive frames of a given hand-drawn animation sequenceis an indispensable tool for automating animation modificationtasks such as sequence-wide recoloring or shape-editing of aspecific animated character. Existing correspondence identificationmethods heavily rely on appearance features, but these featuresalone are insufficient to reliably identify region correspondenceswhen there exist occlusions or when two or more objects sharesimilar appearances. To resolve the above problems, manualassistance is often required. In this paper, we propose anew correspondence identification method which considers bothappearance features and motions of regions in a global manner.We formulate correspondence likelihoods between temporal regionpairs as a network flow graph problem which can be solved by awell-established optimization algorithm. We have evaluated ourmethod with various animation sequences and results show thatour method consistently outperforms the state-of-the-art methodswithout any user guidance.