Beyond Local Appearance Category Recognition from Pairwise Interactions of Simple Features.pdfVIP

Beyond Local Appearance: Category Recognition from Pairwise Interactions of Simple Features Marius Leordeanu1 Martial Hebert1 Rahul Sukthankar2,1 mleordea@ hebert@ rahuls@ 1Carnegie Mellon University 2Intel Research Pittsburgh Abstract We present a discriminative shape-based algorithm for object category localization and recognition. Our method learns object models in a weakly-supervised fashion, with- out requiring the specification of object locations nor pixel masks in the training data. We represent object models as cliques of fully-interconnected parts, exploiting only the pairwise geometric relationships between them. The use of pairwise relationships enables our algorithm to suc- cessfully overcome several problems that are common to previously-published methods. Even though our algorithm can easily incorporate local appearance information from richer features, we purposefully do not use them in or- der to demonstrate that simple geometric relationships can match (or exceed) the performance of state-of-the-art object recognition algorithms. 1. Introduction Object category recognition is very challenging because there is no formal definition of what constitutes an object category. While people largely agree on common, useful categories, it is still not clear which are the objects’ features that help us group them into such categories. Our proposed approach is based on the observation that for a wide variety of common object categories, shape matters more than lo- cal appearance. For example, it is the shape, not the color or texture, that enables a plane to fly, an animal to run or a human hand to manipulate objects. Many categories are defined by their function and it is typically the case that function dictates an object’s shape rather than its low level surface appearance. In this paper we represent these object category models as cliques of very simple features (sparse points and their normals), and focus only on the pairwise geometric relationships betwee