A New Multimodal Biometric for Personal Identification

Abstract

Biometric technology using a single human body characteristic such as face, gait or voice has gained immerse attention with successful applications in video surveillance. Furthermore, facial recognition has been recognized as the least intrusive technology that can be implemented in many places without hazardous problems. Though existing biometric systems have been reported to be effective under certain conditions, there is a great need to improve their recognition performance. Possible alternatives to current approaches include the use of different biometric information or the combination of different biometric sources. Relevant literature indicates the possibility of using facial behavior as another behavior biometric cue. As this biometric cue reflects the internal dynamic changing factors of an individual, it also plays an important role just as the face biometric does in video surveillance. Moreover, existing research in appearance-based facial recognition always addresses the common problem of within-class variations under illumination and poses and/or facial expressions which degrades the recognition performance. Most of the algorithms (Belhumeur et al., 1997) (Chen et al., 2000) (Lu et al., 2003c) (Lu et al., 2003b) (Lu et al., 2003a) (Juwei et al., 2003) (Lu et al., 2005) (Kong et al., 2005) in facial recognition are developed to cope with the singularity problem in the presence of these variations. Some papers (Martinez, 2000) (Liu et al., 2002) (Liu et al., 2003) (Bronstein et al., 2003) (Bronstein et al., 2007) even consider facial expressions as noise that will degrade the system performance, and they attempt to build robust systems that are invariant to these variations. However, no research has been conducted to see if these intra-personal variations, especially under facial expression changes or dynamic changes of intra-personal information could help the extra-personal separation. Can within-class variation help between-class separation? Or can intra-personal facial expression variations assist extra-personal separation? We assume that “intra-personal facial expression variations could assist extrapersonal separation.” In the following section, we give an overview of multimodal biometrics and soft biometrics. We discuss how multiple biometrics and soft biometric can improve classification performance. We then discuss related work that uses other biometrics in combination with facial biometric at a distance and our proposed fusion framework. Finally, we give the experimental results and conclusion. 17