MULTIMODAL BIOMETRIC FUSION

Different methodologies for rank level fusion were presented. These methods included plurality voting method, highest rank method, Borda count method; logistic regression method and quality based rank fusion method for multimodal biometric system. Among those methods, the logistic regression method consistently provides high Performance, however it still has some drawbacks. The results obtained through this method can be varied significantly for different datasets due to their diverse qualities. Logistic regression method for a multimodal dataset with the same image quality will produce results similar to Borda count method, as the assigned weights to different biometric matchers’ outputs will be the same. Thus, allocating appropriate weights to different matchers (comparing different quality datasets) requires appropriate learning technique, which is time consuming. Also, inappropriate weight allocation can result in wrong recognition results. Further, the size of the multimodal biometric database is usually large and thus only the top few results are considered for the final reordered ranking. Hence, a very common scenario of a rank based multimodal biometric system is that some results may rank at top by a few classifiers and the rest of the classifiers do not even output the result. In this situation, the logistic regression approach cannot produce a good recognition performance. Thus, a novel rank fusion method utilizing Markov chain has been recently developed at BT Lab at the University of Calgary. This method can be efficiently used in multimodal biometric authentication system comprised of varied quality datasets. The method has been successfully used in other information fusion applications. In this chapter, an overall description of this method is given. It includes Markov chain definition, advantages and disadvantages of Markov chain in multimodal biometric fusion scenario, previous research on Markov chain and its application in rank level fusion.

                Utilized Markov chain as a method for biometric rank fusion. This approach brought a new dimension to the current ways of biometric rank aggregation and can be effectively used by the homeland and border security forces and by other intelligence services.

                 In their research, they considered the biometric rank aggregation similar to a voting mechanism. In the multimodal biometric rank fusion process, the classifiers are considered as voters. So, if there are three biometric traits used in a multimodal biometric system, the number of voters in the system would be three. Those three voters or classifiers produce three ranking list based on the similarity or distance scores of test and template biometric data. The final process is to combine the ranking lists obtained from three classifiers or voters to make a consensus ranking lists to find out the desired identity or alternative from the system.