From the Mahalanobis distance to the skin color distribution, which was obtained by clustering the color and texture descriptors. All these methods were demonstrated for low-resolution eye images and employed a single shape model for both eye open and eye closed images, so that even when the eye is closed, the iris was still a valid and visible part of the shape model (Orozco et al., 2009), which makes little sense. Tan and Zhang (2006) firstly determined the eye state of open or close by examining the existence of the iris using intensity and edge information. Then the eye feature was modeled as either a straight line or a deformable template before manual initialization and application of a standard Lucas Kanade framework for tracking. Recently, Alabort-i-Medina et al.
Active Appearance Models (AAM), Constrained Local Models (CLM), and Supervised Descent Method (SDM) to track the deformation and motion of eyelid, the iris and pupil. In order to represent the open and fully closed eye correctly, two sets of shape models were used for Spain phone number list the two eye states. In contrast to manual initialization of the shape or texture as in previous work, the initial shape model was statistically learned from training data, making fewer assumptions about the shape and texture of an individual's eye. Learning and evaluation were conducted for open and closed eye images individually. The results demonstrated that for open eye images, AAM performed best when the Cumulative Error Distribution (CED) normalized to eye size was <0.05, otherwise SDM was the best choice.
For closed eye images, SDM performed best. This work demonstrated the potential for applying facial landmark detection approaches, especially the SDM technique, to far-field high resolution eye images for open eye motion tracking, however, their performance for fully closed eye images was poor. The feasibility for near-field IR eye images is unknown, neither of how to determine the open and closed eye states and how the trained shape model affects individual eye alignment whose state is unknown. Blink Detection There are a variety of blink detection techniques in the literature, which can be categorized into four main groups.