Function-based classification from 3D data via generic and symbolic models

We propose a novel scheme for function-based classification of objects in 3D images. The classification process calls for constructing a generic multi-level hierarchical description of object classes in terms of functional components. Functionality is derived from a large set of geometric attributes and relationships between object parts. Initially, the input range data describing each object instance is segmented, each object part is labeled as one of a few possible primitives, and each group of primitive parts is tagged by a functional symbol. Connections between primitive parts and functional parts at the same level in the hierarchy are labeled as well. Then, the generic multi-level hierarchical description of object classes is built using the functionalities of a number of object instances. During classification, a search through a finite graph using a probabilistic fitness measure is performed to find the best assignment of object parts to the functional structures of each class. An object is assigned to a class providing the highest fitness value. The scheme does not require a-priori knowledge about any class. We tested the proposed scheme on a database of about one thousand different 3D objects. The results show high accuracy in classification.