Exploring elitism in genetic algorithms for license plate recognition with michigan-style classifiers

Abstract

This document describes the application of Genetic Algorithms (GAs) in the recognition of the printed characters in Colombian vehicular license plates. First of all, the accuracy achieved by a genetic algorithm with simple elitism is contrasted with the accuracy of a population elitism-based genetic algorithm. Due to the notorious difficulty of using the standard technique of dedicating from, 70 to 80% of the available data to train the classifier, and the rest of data for its validation, here, two methods to generate the training data are described, as well as some other techniques to improve the classifier performance.