Shape Matching and Extraction by an Array of Figure-and-Ground Classifiers

Abstract

For matching a template to a target object in an image under influences from obstructing objects, a two dimensional array of figure-and-ground classifiers is introduced. Each classifier in the array observes a corresponding point in an image and determines if the point belongs to the target object (figure) or its background (ground). Neighboring classifiers communicate via local connections. The local communication is used to transmit the shape transformation parameter values so that the neighboring classifiers interpret their observing points under continuous and topology preserving shape transformation. Some basic experiments were conducted to evaluate the performance of the method and the method’s effectiveness was confirmed.