Multibiometric Authentication

While some biometrics, such as fingerprints, handwritten signatures or facial photographs, have been used in forensics for a good century or longer, modern biometric technology, both for person authentication and for forensic applications, has come to the fore as part of the rapidly expanding capabilities of computer technology in the final decades of the 20th century. A recent review of the history of speaker recognition begins with the words “research in automatic speaker recognition has now spanned four decades” (Furui, 2005), and other biometrics such as face, iris or finger recognition, have been topics for intensive research for about the same period. Only more recently, however, these technologies have moved out of the research labs and into commercial applications, having first required considerable progress in the signal processing and pattern recognition fields to address the difficulties that are inherent in the various biometrics. The main problem in the development of algorithms for the automatic identification or verification of persons is the within-subject variability of the different biometrics. All biometrics show a long-term aging effect, which may be more pronounced in a biometric like face recognition and, for example, less in one like iris recognition. Some biometrics have strong behavioural components and thus significant within-person variation, such as speech, handwritten signature, or facial appearance; and all biometrics are subject to variations in the capture of the person’s biometric features, starting from the environmental conditions, in which the biometric is captured, and the variability of the biometric sensor, through to the distortions in any transmission channel and in the digitization of the analogue sensor data, as shown in Fig. 17.1.