Parameter estimation of non-translational motion fields

Motion estimation is a very important topic in computer vision and image sequence compression. However, most commonly used motion estimation algorithms do not take into consideration any motion invariances that a certain local motion might possess. In this paper, a technique for estimating the invariant motion parameters of non-translational motion fields is proposed, which leads to more efficient estimation, smoothing or coding of the motion field. It is shown that the algorithm performs well, even in high noise levels, i.e., in the case of noisy output of the motion estimator.