Eye State Recognition based on Machine Learning Techniques

Abstract

In this paper we propose a robust approach to determine eye state (open / close) in still images. The method is based on the assumption that an eye patch around the eye center is already found. There are several published works which can accomplish this including our earlier work on automatic eye detection that we will briefly review in this paper. After the eye patch is obtained, Artificial Neural Networks (ANN), Support Vector Machines (SVM) and AdaBoost with different weak learners are combined using a simple majority voting scheme to build a robust classifier for eye state identification. For experiments, a database of 600 images is used which include images with different illumination conditions, taken from different angles and subjects of different ethnicity. Experimental results are shown for each of individual classifier and their combination using majority voting. Self-consistency, Jackknife and independent dataset tests are performed for each individual classifier as well as for classifier obtained using majority voting scheme. When tested by 300 novel eye patches from the database, an accuracy of 100% can be achieved by our proposed majority voting scheme