Combining classifiers using the Dempster-Shafer theory of evidence

As organizations strive for means of providing more secure methods for user access, biometrics is gaining increasing attention. However a biometric recognition system good for one case study may not be accurate for the other one. One solution to the problem is combining classifiers; so that the complementary information departed by different classifiers could be combined, in an efficient way, to achieve a much better recognition rate as compared to the participating experts. In this context the Dempster Shafer theory of evidence (DST) has shown some promising results; however the DST has not yet been explored for the problem of biometric recognition systems. In this thesis we have proposed three novel algorithms to combine different biometric systems using the DST. NNEF (Nearest Neighbor Based Evidence Fusion) algorithm uses the nearest neighbor distance of the participating experts as an evidence estimation parameter; RREF (Recognition Rate Based Evidence Fusion) algorithm uses the performance parameters of the participating experts for evidence estimation and VEF (Variance Based Evidence Fusion) algorithm uses the second order statistics of decision parameters to estimate the belief in the combining experts. (Abstract shortened by UMI.)...