Biometric Fusion

In this chapter we will discuss fusion (i.e., combination) of multiple sources of information with the goal of improving the performance of fingerprint systems. This topic is also known as multibiometrics or multimodal biometrics. Biometric fusion is a very important topic that has already been in use in law enforcement for some time (e.g., fusion of 10 rolled fingerprints in AFIS), but the challenge is to determine which information sources to combine and which combination strategies to use. Answers to these questions are application specific. Our intention is not to make specific recommendation, but rather to discuss the tools available to system designers to make informed decision and implementation. A large body of literature exists on information fusion in many different fields, including data mining, pattern recognition and computer vision. Fusion in biometric (Ross, Nandakumar, and Jain, 2006), and fingerprint systems in particular, is an instance of information fusion (Dasarathy, 1994). A strong theoretical base as well as numerous empirical studies has been documented that support the advantages of fusion in fingerprint systems. The main advantage of fusion in the context of biometrics is an improvement in the overall matching accuracy.

Not surprisingly, the improvement in matching accuracy due to fusion also comes at a cost. For example, the multibiometric system may require many sensors and thus will be more expensive. Further, such a system may also increase the user inconvenience as the user needs to interact with more than one sensor, increasing both the enrollment and verification times. For example, in a multibiometric system that requires both fingerprint and iris images of a person, a user will not only need to touch the fingerprint scanner, but will also need to interact with an iris imaging system. Such a system may also require more computational power and storage.