Two‐dimensional gray‐level object recognition using shape‐specific points

Abstract

This paper defines the nth order radius weighted mean points (RWMP) where n is any integer, and shows that they are shape‐specific with respect to a 2‐D gray‐level object under translation, rotation and scaling. The distances between these RWMP form a feature vector that is invariant to translation and rotation. To achieve the scale invariance, we can normalize the feature vector by dividing it by the distance between centroid and first order RWMP. The Euclidean distances between the normalized feature vectors of the input and reference objects are used as a dissimilarity measure to recognize the objects.