FUSION AFTER MATCHING BIOMETRICS

There are a number of challenges in this area, requiring further investigation. The first one is rooted in the choice of a fusion method, most appropriate for the application domain. The decision is often made ad-hoc, or based on non-essential constraints such as availability of the fusion module, low cost, etc, instead of being made based on actual fit of the application area and the method.

Arguably, one of the critical components of the multimodal biometric system development is an information fusion module. It is also a component which is most versatile in the form of input data (processed or unprocessed), types of features (geometric, signal, appearance-based, etc), and decision making process (adaptive, intelligent, fuzzy, learning-based, heuristic-based) it can utilize. Needless to say, the initial choice of biometric—physical, behavioral, soft, or social would both be an input to the information fusion process and dictate some of the choices to be made.

A general rule in theory assumes that the integration of data at an early stage of processing leads to systems which might be more accurate than those where the integration is introduced at later stages. Unfortunately, in practice, fusion at sensor level is hard to achieve, due to the different natures of the biometric traits, which might be hardly compatible (e.g., fingerprint and face). Moreover, most commercial biometric systems do not provide access to the feature sets vanishing the feasibility of a fusion at feature level. Fusions at matching level and at decision level do not require the creation of new databases or matching modules (the ones which constitute the mono modal subsystems are employed).

The rank level fusion approach is used in biometric identification systems when the individual matcher’s output is a ranking of the “candidates” in the template database sorted in a decreasing order of match scores (or, an increasing order of distance score in appropriate cases). The system is expected to assign a higher rank to a template that is more similar to the query. Plurality voting method, highest rank method, Borda count method, logistic regression method, Bayesian method and quality based method are reported in the literature to perform rank level fusion in multi biometric system. All of these biometric rank fusion approaches.