Shape-based object recognition by inductive learning

Describes a system that learns to classify objects based on descriptions of their shape. A shape given by line drawings is represented by conjunctions of local properties (CLP). The line drawings are approximated by straight line segments and local properties are computed at each segment. An inductive learning method called property-based learning (PBL) is developed to learn to classify shapes represented by CLP. The PBL indexes objects by the properties of an instance and matches the indexed objects against the instance by a similarity model. The PBL learns the information necessary to do indexing and matching. Experimental results applying these techniques to the task of classifying different tools are presented