Facial Image Recognition Using Neural Networks and Genetic Algorithms

This paper addresses the design of a simple yet efficient facial image recognition system. We show that a face can be recognised based on the relative size and position of its basic features, i.e., eyes, nose and lips. The key to the efficiency of our algorithm is in the feature search method employed. Feature search is accomplished through the combination of conventional template matching and genetic algorithms. Genetic algorithms alone would take a long time in order to converge to a valid solution. However, by first performing a coarse but fast template matching, we can obtain an approximate solution that can be utilised to initialise the genetic algorithm. The output of the facial feature detection stage is fed to a back-propagation neural network which accomplishes the recognition task. Our experimental results show that the system is very efficient both computationally and in recognition accuracy as long as the facial database to be recognised has a moderate size (16 in our experiments). We also note that the basic ideas conveyed in this work can be easily generalised to general purpose object recognition applications.