TOPOLOGY-BASED INTELLIGENT PATTERN ANALYSIS IN BIOMETRICS

The optimal biometric system is one having the properties of distinctiveness, universality, permanence, acceptability, collectability, and security. As we saw in the introductory chapters, no existing biometric security system simultaneously meets all of these requirements. Despite tremendous progress in the field, over the last decades researchers noticed that while a single biometric trait might not always satisfy secure system requirements, the combination of traits from different biometrics will do the job. The key is in aggregation of data and intelligent decision making based on responses received from individual (unimodal) biometric systems.

Thus, Multimodal biometrics emerged as a new and highly promising approach to biometric knowledge representation, which strives to overcome problems of individual biometric matchers by consolidating the evidence presented by multiple biometric traits. As an example, a multimodal system may use both face recognition and signature to authenticate a person. Due to reliable and efficient security solutions in the security critical applications, multimodal biometric systems have evolved over last decade as a viable alternative to the traditional unimodal security systems.

The goal of any intelligent processing is to minimize overhead associated with performing computations while at the same time to maximize an output. The same principle governs behavior of most public and commercial organizations—to achieve high production by resource and processes optimization. While appearance-based methods excel in capturing even subtle features in the multitude of high-dimensional data, sometimes generalizing the results and noting common patterns leads to process optimization without sacrificing the security system performance. This section presents topology-based methods, which work best with biometric data that has prominent geometric features, such as fingerprint or hand/palm biometrics. We start by outlining the topology-based methodology with the roots in computational geometry.