Video coding using image warping within variable size blocks

Current video coding standards use simple translational motion models in predefined blocks to reduce temporal redundancy. Although this approach is quite effective, artifacts are produced by the simplistic motion model used and the fixed motion boundaries. The codec presented in this paper removes the constraint of fixed block size and translational motion to allow more flexibility in the motion field. We derive a new algorithm that directly computes the coefficients for representing a motion field modeled by a set of arbitrary basis functions. The motion field model is then used to predict blocks in the current frame by warping blocks in the previously coded frame. This motion compensation algorithm is incorporated into a transform based coder and compared to an existing H.263 codec. Results show up to 10% savings in bit rate for the same quality.