Analysis of different threshold selection methods for eye image segmentation used in eye tracking applications

The purpose of this paper is to compare two established threshold selection methods used for eye image segmentation: quantitative fixed threshold and Kittler minimum error threshold with a new one based on characteristic selection adaptive threshold. This method relies on identifying the characteristic that describes the pixels that belong to the eye pupil. The results are obtained by using as reference the real pupil contour that was manually obtained by analyzing a number of 52 infrared images captured from a set of healthy individuals in laboratory condition.