Invariant quantities in regression-induced boundaries under a special linear transformation

摘要

Boundaries for most objects are not in simple mathematical forms, but they can be approximated by simple curves through regression. This report shows that invariant quantities can be found when boundaries of an object and its image are approximated by regression curves. The estimations of these invariants and their statistical properties are derived and a method to combine the information in a sequence of pictures is also discussed. The classification of motion-induced radar ship imagery is used as an example.