Efficient eye detection method based on grey intensity variance and independent components analysis

Detection of facial features such as eye, nose and mouth in the human face images is important for many applications like face identification or recognition systems. Independent components analysis (ICA) is an unsupervised learning method which decorrelates the higher-order statistics in addition to the second-order moments. Recently, it is used as a technique for face recognition. In this study, ICA applied on a patch image is used as a method to extract the eye which is the most salient and stable feature among all the facial features. The variance of grey intensity in the eye region and ICA are combined together to detect rough eye window. The ICA basis images are computed using the FastICA algorithm; that computes independent components by maximising non-Gaussianity of the whitened data distribution using a kurtosis maximisation process. After detecting rough eye window, intensity information is used to localise eye centre point. The proposed method is evaluated on different databases XM2VTS, BioID and FERET and experimental results demonstrate improved performance over the existing methods. In addition, a high detection rate of 93.3% can be achieved on 600 images with glasses. A comparison between the proposed method and the most recent published methods which focus on eye window detection is also reported.