LEARNING RELATIONAL STRUCTURES - APPLICATIONS IN COMPUTER VISION

We present and compare two new techniques for learning Relational Stru ctures (RSs) as they occur in 2D pattern and 3D object recognition. Th ese techniques, namely, Evidence-Based Networks (EBS-NNets) and Rulegr aphs combine techniques from computer vision with those from machine l earning and graph matching. The EBS-NNet has the ability to generalize pattern rules from training instances in terms of bounds on both unar y (single part) and binary (part relation) numerical features. It also learns the compatibilities between unary and binary feature states in defining different pattern classes. Rulegraphs check this compatibili ty between unary and binary rules by combining evidence theory with gr aph theory. The two systems are tested and compared using a number of different pattern and object recognition problems.