ASSOCIATIVE STRUCTURES FOR VISION

The paper describes how associative techniques can contribute to reduc ing data dimensionality for image understanding. Associative memories enhance a vision system's robustness by their pattern-completion capab ility; neural networks compensate for the limited storage capacity of the memory arid provide adaptive nonlinear filtering to remove crossta lk noise. Two associative schemata are defined: for visual classificat ion and for stimulus response. The major concern is to preserve genera l applicability, while ensuring practical effectiveness. Both a theore tical discussion and experimental evidence support the satisfactory re sults obtained.