İçbükey eşleştirmeleriyle nesne tanıma

Abstract

In this work, we present a new method for recognizing planar objects by matching multiple concavities of object boundaries in a projective invariant manner. Object boundaries are first decomposed into several concavities using bitangents and obtained concavities are then sorted. Corresponding concavities are aligned by using the projective matrix computed from their bitangent and tangent points. Best concavities are selected based on the Hausdorff distance between aligned concavities. This subset of concavities is then used to align original and transformed objects. Closeness of two aligned objects is checked by the same distance metric. Proposed method is verified by several experiments.