Motion texture for dynamic sequence synthesis

We describe a novel model, called motion texture, for synthesizing complex dynamic sequences that are statistically similar to the original sample data. We define motion texture as a set of motion textons and their distribution, which characterize the stochastic and dynamic nature of the sample data. Specifically, a motion texton is modeled by a linear dynamic system (LDS) while the texton distribution is represented by a transition matrix indicating how likely each texton is switched to another. We design a maximum likelihood algorithm to learn the motion textons and their relationship. The learnt motion texture can then be used to generate new animations automatically. Our approach is demonstrated by many synthesized sequences of visually compelling dance motion and video sequences.