Abstract
The COUGAR project (an EC-funded project within the RACE programme) is studying the use of high-quality motion estimation methods to provide vectors and other control information for the MPEG-2 coder being developed in the project. It is hoped that the use of a motion estimator, optimized to produce `true' motion vectors, will lead to higher quality pictures compared to the use of a simple `best match' motion estimator. This may be particularly significant in the presence of artefacts such as noise, grain and unsteadiness. The approach is driven by the observation that it is possible to perform high-quality motion-compensated temporal interpolation (for example, in standards converters and advanced slow motion generators) and obtain excellent results, without the use of any prediction error signal to correct errors in the interpolation.