The researchers are studying motion estimation for applications in image processing such as motion compensation in image encoding. The researchers are also using motion estimation for improving image sequence restoration and in the problem of interpolating between two adjacent image frames. They have developed a new approach to motion estimation which is model- based and uses a multi-grid for computational efficiency. The method involves a careful solution of two linear velocity equations resulting from a least-squares analysis. Also for noise immunity, they approximate the relevant partial derivatives by low order polynomial interpolation rather than first order differences. The new technique includes an accurate detection of regions where the motion estimates are not good enough to be useful for the above tasks.
