Robust method for Gaze Recognition Using HistogramThresholding and Neural Network

The existing studies discuss an effective method for individual’sgaze recognition in unconstrained environments. In first step,coordinates of pupil center have been calculated for gazerecognition. In suggested method and algorithms, firstly, pupilcenter have been localized by histogram Threshold and AverageFilter. Some steps of algorithms have been optimized in thepurpose of reducing complexity, performance optimization, andincreasing the speed of algorithms. At first, by applying themethod of Connected Component Analysis, the pupil center andthen the peripheral circle of the edge point of pupil and Iris havebeen localized. Datasets MMU1, MMU2, and UBIRIS3 havebeen utilized for testing this method. In next step, dataclassification in gaze recognition system is performed byapplying multilayer neural network and the Learning algorithmof 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 iscalculated and our final result is 88%.

کلیدواژه ها:

Gaze recognition ، histogram thresholding ، Average filter ، neural network ، Back- propagation