Fusing correlated data from multiple classifiers for improved biometric verification

Dynamically weighting and combining data from correlated biometric classifiers, having different statistical distributions and characteristics, is accomplished via a Particle Swarm Optimization algorithm. Real time data collected from correlated biometric classifiers are fused using various normalizing score fusion techniques, z-normalization and min-max normalization. Since individual classifiers have varying degree of accuracy, weighting is paramount to achieve higher benefits. Weights are found using a PSO algorithm and are a function of accuracy and degree of correlation. Results are presented for, a) synthetic score data generated using a multivariate normal distribution with different covariance matrices and b) NIST BSSR dataset. The performance of the PSO technique, for correlated biometric classifier, is better than the traditional score level fusion techniques.