Weiss describes a method for object representation and recognition which includes a semantic and geometric understanding of the objects. One benefit is avoiding the pitfalls of current popular methods (machine learning) which train a system on large numbers of images but without any “deeper” understanding comparable to human intelligence. Weiss represents a data-set of objects by one dual-hierarchy graph which incorporates all our structural, geometric and semantic knowledge about the objects, as well as their mutual relations. The representation is invariant to parameters such as view-point. Recognition is performed by traversing both hierarchies in the graph simultaneously.