An Abstraction Based Model of Object Classification

This document discusses an abstraction-based model of object recognition. The abstraction model provides a representation of contour data that is invariant of isomorphism. The constraint that the contour globally represents the object is lifted, and the recognition problem is approached locally. It is shown that the local perspective allows for the treatment of object recognition in the presence of partial or incomplete contour information. It is observed that this enables capabilities such as recognizing an object in partial occlusion, and the recognition of objects that have been locally deformed