Globally Optimal Toon Tracking
Paper
Globally Optimal Toon Tracking
ACM Transactions on Graphics (SIGGRAPH 2016 issue)
Problem
Tracking object and region correspondences across frames in hand-drawn animation is difficult when appearance cues are ambiguous, objects undergo occlusion, or multiple regions share similar visual characteristics.
Method
This work formulates temporal region correspondence as a global optimization problem over both appearance and motion cues. The correspondence likelihoods are modeled as a network-flow graph and solved with a global optimization algorithm.
Results
The method was evaluated on multiple animation sequences and showed consistent improvements over prior correspondence methods without requiring user guidance.