Face recognition is a complex task which has recieved a great amount of attention in recent years, mostly due to its wide range of application in the area of biometric systems. Many different approaches have been proposed, and current systems exhibit very good performance on detection, identification and verification of human faces [10,2]. However, one of the remaining problems is dealing with changes in faces over time. Since eyes are one feature of the face which are not greatly affected by certain typical face changes (e.g. facial hair like beard or mustasch), we address the problem of identifying subjects from the eyes only.
There is a significant amount of evidence from psychophysics which support the theory that humans make use of local features (such as the eyes) when recognizing other individuals [1]. And indeed, several current face recognition systems are based on local features as well as global (holistic) features. The eigenface approach, developed by Turk et al. [12], was further developed in [8] to include eigenfeatures such as eigeneyes, eigennose and eigenmouth. This improved performance of the system, and result were also surprisingly good when only the eigenfeatures were used. A very different system for object recogniton was developed by Lades et al. [5] and applied to face recognition in [13,9]. This system is based entirely on local features, computed by a biologically motivated [3] wavelet transform. These systems are two of the best face recogniton systems today, with different advantages/disadvantages.
In this paper we investigate how well human faces can be recognized when the features are extracted from the eyes only. We explore performance with the eigenfeature technique [8], which is applicable for both local (eyes, nose, mouth, etc.) and global (entire face) feature extraction, and the gabor wavelet approach [5], which is a typical local feature extractor. We compare these techniques with classification directly from gray-level values. Recognizing faces from just the eyes is important work, because in many situation the eyes can be the most reliable (and perhaps the only) source of information. Our experimental results show surprisingly good performance on a small, but difficult dataset.