PARALLEL IMPLEMENTATION OF ASSOCIATIVE MEMORIES FOR IMAGE CLASSIFICATION

Associative techniques are useful in computer vision because they are notably able to robustify a recognition system. The noise-like coding model of associative memory has been already applied successfully to i mage-classification. This paper describes the implementation of the as sociative system on transputer-based architectures. After explaining t he model's basic formalism, the paper marks out the key-generation mec hanism, the data-mapping strategy, and the hierarchical processor orga nization. The basic result of this research is a general methodology f or efficient HW configuration of real-time associative visual systems. The system's efficiency can be predicted by theoretical derivations, in which both the FFT-computation speed and the data-transmission spee d play a crucial role. Experimental results including different HW con figuration and different image-sizes always confirmed theoretical expe ctations.