LINEAR FEATURE COMPATIBILITY FOR PATTERN-MATCHING RELAXATION

This paper extends point-pattern matching relaxation techniques to allow matching of both point-like and linear features. We define a compatibility function that relies on relative orientation information, which is translation and rotation invariant and which can be more reliably extracted from noisy images than can positional information. The function is used to generalize the matching technique of Ranade and Rosenfeld (Pattern Recognition 12, 269-275 (1980)), but it can also be incorporated into other relaxation algorithms. Examples are presented from the domain of object recognition in synthetic aperture radar (SAR) imagery.