Fast half-pixel motion estimation using steepest descent

摘要

A method of locating a half-pixel resolution motion vector. The method first determines a sum of absolute difference for a predetermined number of neighbors to a current pixel. The gradient of these sums of absolute differences is then calculated. A motion vector is then selected based upon the gradient that indicates which vector lies in a direction of steepest descent. This motion vector is then used in motion compensation for video compression.