Enhanced video coding based on camera motion and image synthesis

In this paper, we propose a method that efficiently models video transitions and video effects (e.g. fade, zoom) present in video sequences for video coding purposes. In the proposed scheme, the transition model parameters are extracted at the encoder and then delivered to the decoder side. The decoder includes an image synthesizer able to regenerate the transition video frames and results in considerable coding gain while keeping an acceptable reconstruction video quality. Our method currently employs camera motion estimation and image morphing to model transitions such as camera zoom-in and zoom-out. We apply our method to stationary camera zoom shots and regenerate visually similar sequences to the original ones. The quality of the regenerated video is evaluated by color texture and motion similarities.