Iris Region and Bayes Classifier for Robust Open or Closed Eye Detection

Abstract

Abstract

This paper presents a robust method to detect sequence of state open or closed of eye in low-resolution image which can finally lead to efficient eye blink detection for practical use. Eye states and eye blink detection play an important role in human-computer interaction (HCI) systems. Eye blinks can be used as communication method for people with severe disability providing an alternate input modality to control a computer or as detection method for a driver’s drowsiness. The proposed approach is based on an analysis of eye and skin in eye region image. Evidently, the iris and sclera regions increase as a person opens an eye and decrease while an eye is closing. In particular, the distributions of these eye components, during each eye state, form a bell-like shape. By using color tone differences, the iris and sclera regions can be extracted from the skin. Next, a naive Bayes classifier effectively classifies the eye states. Further, a study also shows that iris region as a feature gives better detection rate over sclera region as a feature. The approach works online with low-resolution image and in typical lighting conditions. It was successfully tested in  image sequences (  frames) and achieved high accuracy of over  for open eye and over  for closed eye compared to the ground truth. In particular, it improves almost  in terms of open eye state detection compared to a recent commonly used approach, template matching algorithm.