APPLICATIONS OF A CMOS ANALOG CNN TO FILTERING OPERATIONS FOR VISUAL PROCESSING
Neural network, CNN, CMA, Genetic Algorithm, Image Filtering
This work is a contribution to the realization in CMOS technology of an analog circuit, belonging to the biomorphic class of cellular neural networks, for image processing. Aiming at the reproduction of more complex operations than usual, a compact cell architecture was employed in the implementation of a network featuring two layers and mutual coupling, originated from an extension of the traditional CNN. To obtain the parameters that configure the network to perform the desired functions, a training methodology was developed, taking inspiration from some existing techniques addressed to this category of neuronal network and including modifications either to improve its performance in general or to adapt the process to some relevant characteristics of the various types of image operations considered. For verification purposes, the circuit is applied in simulations involving bipolar functions and image filtering, exhibiting similar results to the ones generated from an ideal model.