View article

An object’s shape is often the first visual cue we exploit for recognition. In fact, in many cases shape dominates any other appearance feature. A red elephant or a spotless giraffe, for example, can be correctly recognized despite the unusual appearance. In addition to its category, an object’s shape can recover a great deal about the objects, like its activity, pose, and relations to other objects. For example, the shapes depicted in Fig. 1.1 reveal the sport activity that each object is performing. Nevertheless, as simple as it is for our visual system to recognize and classify objects on the basis of their shape, an artificial system with similar performance is still out of reach. In this thesis we present several algorithms for shape-based recognition. We present a shape similarity measure for comparing clean segmented objects. The proposed similarity measure generalizes the definition of the term “shape” as given by Kendall …