Even though numerous techniques for face recognition have been explored over the years, most research has primarily focused on identification from full frontal/profile facial images. This paper conducts the first systemic study to assess the performance when using partial faces for identification. In this study, two partial face recognition approaches have been proposed. In the first method, artificial neural network is used as the classifier for multi-scaled face components such as eyes, nose and mouth. While in the second method, eigen features/linear discrimination features of face components are used as inputs for artificial neural network. In both methods, it is found that with partial face (face components), the percentage of recognition drops only slightly, compared to the full face recognition, and that eyes play a significant role in recognition. Experiments are conducted on ORL and FERET databases.

One of the most important objectives of computer vision is to mimic human performance in rapid classification of complex and unique images such as human face. Face recognition has become a fundamental research topic in computer vision, with a large number of studies emphasizing on the emerging security threats such as terrorism and organized crime. Conventional applications of face recognition system, especially those of surveillance, access control and authentication, typically require maximum information of face to achieve good recognition performance. However, sometimes, a full face cannot be obtained under certain restricted circumstances. For example, a noncooperative face can lead to face occlusion. Moreover, people may be required to hide their face due to occupation (i.e., soldier), contagious disease (i.e., SARS), scars or religious practices. Figure 1 shows a Muslim woman wearing a veil. The issue of Muslim women wearing veils in public must be considered in the research of face recognition. Thus, the partial face could be one of the potential solutions to solve the restricted circumstances mentioned above.

In the context of this paper, the term `partial' refers to a single part or a combination of parts of frontal face, such as eye, nose and mouth. Recently, a number of partial face recognition works (Gutta and Wechsler, 2002; Gutta et al., 2003; Tarres and Rama, 2005; Oh et al., 2006; and Savvides et al., 2006) have been evaluated. Savvides et al. (2006) have reported that human eye is a part of the face that provides the best discrimination information. Gutta et al. (2003) showed that there is no significant difference between the partial face and full face. However, the difference between a smaller part of face such as eye, nose and mouth and the full face has not been evaluated. Motivated by Gutta et al. (2003), and Savvides et al. (2006), and to overcome face recognition under the restricted circumstances mentioned above, we use human eye, nose and mouth as the parts of face for further recognition.

Electrical and Electronics Engineering Journal, Partial Face Recognition, FERET Databases, Conventional Applications, Face Recognition System, Artificial Neural Network, Linear Discriminant Analysis, Principal Component Analysis, Neural Processing Phases, Equalization Process, Histogram Equalization, Vector Machine Kernel.