Further Study of Digital Matching of Dissimilar Images.

摘要

This report reviews the L.N.K. registration algorithm and provides an example. The matching of radar and optical imagery was investigated. Special processing of the radar imagery was found to be required and an edge-preserving local filter was successfully applied. A foundation was begun for transformation reliability estimation through the concept of point pattern spaces, as an alternative preferable to least squares. Pattern spaces for translation, rotation, and point reordering were considered. Keywords include: edge detection; filtering; match accuracy; point feature matching; and registration.