MOTION FIELD MODELING FOR VIDEO SEQUENCES

In this paper, we propose a model for the interframe correspondences e xisting between pixels of an image sequence. These correspondences for m the elements of a held called the motion field, In our model, spatia l neighborhoods of motion elements are related based on a generalizati on of autoregressive (AR) modeling of time-series, We also propose a j oint spatiotemporal model by including spatial neighborhoods of pixel intensities in the motion model, A fundamental difference of our appro ach with most previous approaches to modeling motion is in basing our model on concepts from statistical signal processing, The developments in this paper give rise to the promise of extending well-understood t ools of signal processing (e.g., filtering) to the analysis and proces sing of motion fields. Simulation results presented show the excellent performance of our models in interframe prediction; specifically, on average the motion model performs 29% better in terms of mean squared error energy over a commonly used pel-recursive approach [1], The spat iotemporal model improves prediction efficiencies by 8% over the motio n model, Our model can also be used to obtain estimates of the optical flow field as simulations will demonstrate.