Shape Recognition Based on an a Contrario Methodology

Pablo Musé, Frédéric Sur, Frédéric Cao, Yann Gousseau, Jean-Michel Morel

Abstract: The Achilles’ heel of most shape recognition systems is the decision stage, whose goal is to clearly answer the question of whether two shapes look alike or not. In this chapter we propose a method to address this issue, that consists in pairing two shapes as soon as their proximity is unlikely to be observed “by chance.” This is achieved by bounding the number of false matches between a query shape and shapes from the database. The same statistical principle is used to extract relevant shape elements from images, yielding a complete procedure to decide whether or not two images share some common shapes.