Motion estimation and image warping for video compression

Image warping is commonly seen on television as a special effect, bending an image into a cylinder or such like. But image warping can also have uses in video compression. As future trends head towards higher quality digital broadcast television and increased service traffic on transmission networks the need for levels of higher compression and fidelity is driving research into techniques beyond MPEG-2. These second generation coding techniques are aiming for a human visual model approach where objects in an image sequence are recognised as cohesive, but perhaps flexible, units. The motion of these objects and their apparent deformation in shape can be estimated and characterised by a number of descriptive parameters, which can then be transmitted. At the receiving end, a reconstruction of the moving object can be obtained from the object in a previous frame by using the parameters to define the necessary motion and deformation. This approach to motion estimation/compensation allows for a more natural movement of the object than purely translational models. The paper describes the attempts to produce a minimal but sufficient estimate of the motion of a flexible object. It is contrasted with traditional block matching techniques, and the improvements in the prediction image obtained with warping motion estimation and compensation are shown