Implementation of a High Accuracy Pupil Detection Algorithm Using Neural Networks

A high-accuracy pupil detection algorithm implemented using neural networks is presented. The technique used is that of a classifier, the classes being defined according to the possible positions of the pupil center. In order to reduce the number of neurons and therefore the complexity of the network, a parallel architecture was used, the two coordinates of the pupil center being delivered by different networks. Training, testing and validation were done using eye images from three databases previously used in testing algorithms based on traditional image processing.