COMPUTING CLOUD MOTION USING A CORRELATION RELAXATION ALGORITHM — Improving Estimation by Exploiting Problem Knowledge

A Correlation Relaxation (C-R) algorithm for computing wind velocities from satellite images was recently developed by the author (Wu 1993). Unlike the traditional maximum cross correlation (MCC) method in which only the MCC position (MCCP) in a correlation coefficient matrix is used to derive a displacement, the new method considers many positions associated with high correlations in the matrix, including the MCCP, as candidates for the displacement. It selects an estimate from all candidates using a relaxation labeling technique which iteratively updates the likelihood of each candidate according to a few constraints expressing relevant problem knowledge. The C-R method is shown in this paper to have the potential to exploit a wide range of problem knowledge for improving cloud motion estimation from sequential satellite images. Further developing the C-R method is also justified on the need for more automated estimation of wind vectors in order to reduce or even eliminate manual editing work. In addition, the C-R method can create denser wind vector field than the MCC method.