Robust method for Gaze Recognition Using Histogram Thresholding and Neural Network

Abstract

The existing studies discuss an effective method for individual`s gaze recognition in unconstrained environments. In first step, coordinates of pupil center have been calculated for gaze recognition. In suggested method and algorithms, firstly, pupil center have been localized by histogram Threshold and Average Filter. Some steps of algorithms have been optimized in the purpose of reducing complexity, performance optimization, and increasing the speed of algorithms. At first, by applying the method of Connected Component Analysis, the pupil center and then the peripheral circle of the edge point of pupil and Iris have been localized. Datasets MMU1, MMU2, and UBIRIS3 have been utilized for testing this method. In next step, data classification in gaze recognition system is performed by applying multilayer neural network and the Learning algorithm of back-propagation of errors .For testing suggested method, around 234 images of Colombia Gaze Datasets have been used. The performance of this method for our mentioned dataset is calculated and our final result is 88%.