Towards Explaining the Success (Or Failure) of Fusion in Biometric Authentication

Combining multiple information sources, typically from several data streams is a very promising approach, both in experiments and to some extents in various real-life applications. A system that uses more than one behavioural and physiological characteristics to verify whether a person is who he/she claims to be is called a multimodal biometric authentication system. Due to lack of large true multimodal biometric datasets, the biometric trait of a user from a database is often combined with another different biometric trait of yet another user, thus creating a so-called a chimeric user. In the literature, this practice is justified based on the fact that the underlying biometric traits to be combined are assumed to be independent of each other given the user. To the best of our knowledge, there is no literature that approves or disapproves such practice. We study this topic from two aspects: 1) by clarifying the mentioned independence assumption and 2) by constructing a pool of chimeric users from a pool of true modality matched users (or simply “true users”) taken from a bimodal database, such that the performance variability due to chimeric user can be compared with that due to true users. The experimental results suggest that for a large proportion of the experiments, such practice is indeed questionable. Biometric authentication is a process of verifying an identity claim using a person’s behavioral and physiological characteristics. Due to vulnerability of the system to environmental noise and variation caused by the user, fusion of several biometric-enabled systems is identified as a promising solution. In the literature, various fixed rules (e.g. min, max, median, mean) and trainable classifiers (e.g. linear combination of scores or weighted sum) are used to combine the scores of several base-systems. Despite many empirical experiments being reported in the literature, few works are targeted at studying a wide range of factors that can affect the fusion performance. Some of these factors are: 1) dependency among features to be combined, 2) the choice of fusion classifier/operator, 3) the choice of decision threshold, 4) the relative base-system performance, 5) the presence of noise (or the degree of robustness of classifiers to noise), and 6) the type of classifier output. To understand these factors, we propose to model Equal Error Rate (EER), a commonly used performance measure in biometric authentication. Tackling factors 1–5 implies that the use of class conditional Gaussian distribution is imperative, at least to begin with. When the class conditional scores (client or impostor) to be combined are based on a multivariate Gaussian, factors 1, 3, 4 and 5 can be readily modeled. The challenge now lies in establishing the missing link between EER and the fusion classifier mentioned above. Based on the EER framework, we can even derive such missing link with non-linear fusion classifiers, a proposal that, to the best of our knowledge, has not been investigated before. The principal difference between the theoretical EER model proposed here and previous studies in this direction is that scores are considered log-likelihood ratios (of client versus impostor) and the decision threshold is considered a prior (or log-prior ratio). In the previous studies, scores are considered posterior probabilities whereby the role of adjustable threshold as a prior adjustment parameter is somewhat less emphasized. Several issues which are untreated in the EER models are also discussed and supported by some 1186 experiments. These issues are 1) what if the scores are known to be not approximately normally distributed (for instance those due to Multi-Layer Perceptron outputs); 2) what if scores among classifiers to be combined are not comparable in range (their distributions are different from each other); 3) how to evaluate the performance measure other than EER using the proposed EER models.