Knowledge representation for vision: an associative network for single object representation and recognition

VIKNET (visual knowledge network), an extension of associative networks that is specifically designed for knowledge representation for recognition of 3-D objects, is introduced. The design criteria, largely influenced by early vision processing, are discussed. VIKNET is formally described as an algebraic structure, with network functions defined on it. The input image, also in the form of a network, is recognized by a partial matching algorithm. A completely worked out example demonstrates the efficacy of the scheme even for shapes subjected to viewing angle transforms.<<ETX>>