Switching Model-Based Sub-Pixel Motion Estimation

The error criterion function of motion estimation is well represented by mathematical models. The model approach is attractable for real-time application since it reduces dramatically the number of computations compared to conventional exhaustive methods. Its coding performance gets affected by the model accuracy. Here model parameters are computed from the error values of surrounding integer pixels with one pixel distance. Since the derived model can be applied to the surrounding pixels of two pixels distance, the difference between the actual values and the estimated from models are used as a metric to choose the best between models. Through experiments carried out in the h.263 framework, it has been proven that the proposed method dynamically switching between two models can outperform other two methods up to 10% in terms of relative bit rate, while neither of two methods performs the best all the time.