MSR has a paper describing a very clever trick for video. They describe it for automatically creating cartoons, but the applications extend beyond that. Basically, one of the things those clever kids have been paying attention to is looking at video as a 3-D structure and applying algorithms to the resulting “shapes”. In this case, you lay down reference points on two frames, and they tween the reference points not by linearly interpolating between X and Y, but by creating a spline that follows the defined color/shape as it goes through Z (time).