Experimental car plate recognition system by neural networks and image processing

In this paper a system for Italian-style car license plate recognition is described, based on the use of general purpose feedforward neural networks and basic image processing algorithms. The neural network replaces the entire pattern matching module encountered in similar systems. The algorithm actually does not make use of large databases, and experiments were conducted by training the neural classifier on a purely artificial training set. Nevertheless, final results are similar to those of more complicated approaches, and could be further improved by on line learning. The use of neural networks, coupled with the fast BRLS training algorithm, can substantially enhance the global recognition speed.