Abstract
This thesis focuses on developing an algorithm to the detection of the motion of closed-eye using image processing methods. The designed system is able to determine whether the input frame of eye image is a frame with eye-motion or without it. In the procedure, first a preprocessing stage is performed. The preprocessing step is to confine the input image to a certain region of interest including only the eye area. Second, the region of interest is used to extract three different types of features: spatial domain features, statistical features, and frequency domain features. These features are adopted to make decisions of the moving or freezing state of the eye, by a classifier of the Support Vector Machine (SVM) of Neural Networks. The effectiveness of the three types of features and their combinations are analyzed and the correct rates are compared. In addition, two different methods of experiments are attempted: the leave-one-out and the leave-one-person-out methods. The former is to use the prior knowledge of the examinee himself as the training data, while the latter is to use the database collect by us to examine a new examinee’s eye motion. The best correct rate for the leave-one-out method is 95.5% using the spatial features for examinee D, while the best correct rate for the leave-one-person-out method is 96.0% using the spatial and statistical features for examinee G. The results are quite satisfactory and can make contributions to the sleeping monitoring clinical research.