A cellular neural network embedded in a dual computing structure (CNND) and its use for character recognition

Analog neural networks and various regular analog processing arrays are superior in speed, in various tasks, to digital-logical machines. However, in real problems logical decisions and control are generally inevitable. The combination of analog neural and digital logical systems in a dual computing structure has been proposed by T. Roska (1987). A special type of dual computing structure is proposed as a system for character or symbol recognition purposes. It contains a cellular neural network for real-time feature extraction. This cellular neural network embedded in a dual computing structure (CNND) provides a framework for designing algorithms containing sequences of analog and digital operations. This CNND architecture is then used for some well-defined character recognition tasks. Mainly the recognition of printed characters is studied. Several dual algorithms are introduced for classifying various classes of printed character sets. These algorithms contain analog CNN cloning templates and digital decision functions. Preliminary experimental results are presented