Systolic implementation of higher-order CMAC and its application incolour calibration

摘要

The primary advantages of a CMAC neural network are fast learningnand insensitivity to the order in which training patterns are presented.nThe authors present an extended direct weight cell address mappingnmechanism based on a linear systolic array architecture to realise anhigher-order CMAC neural network with digital hardware. Thisnhigher-order CMAC implementation has been applied to calibrate andncompensate the nonlinearity of chromatic mapping between colour scanningnand printing devices in a colour image reproduction environment. A 20nMHz prototyped CMAC chip for colour calibration has been implemented tonconfirm the proposed design approach. Using this prototype, the authorsnwere able to achieve reproduced colour images with rich and vividncolours which strongly resemble the original