Analysis of the effects of image transformation, template selection, and partial information on face recognition with time-varying expressions for homeland security applications

Facial recognition, especially with time-varying facial expressions and/or disguises, is crucial in many homeland security applications. The recent Boston Marathon attack is one example reminder of the importance of developing accurate and reliable facial recognition algorithms. While various face recognition algorithms have been proposed in the literature, unfortunately many of them still remain in their infancy. This is mainly due to their lack of high recognition rates in the presence of varying image face artifacts and conditions. In order to develop more accurate facial recognition systems there is a primary need to identify and, as much as possible, derive some of the causes that may affect some face recognition accuracy rates. The main contribution of this paper is the investigation and analysis of how and what factors, other than illumination noise, and occlusion, may affect the recognition accuracy rate of some of the most popular and currently widely used face recognition algorithms, namely, Eigenface-based, Fisherface-based and Direct Correlation-based ones. In particular, in this work we show the effects, on these facial recognition accuracy, of facial reasonable registration with or without off-the-plane face rotation, the type and number of individual's face template(s) selection, and the type and increasing amount of partial facial information contained in face images. Finally experimental results are presented to demonstrate the potential value and importance of each of these proposed factors on facial recognition.