DS theory based fingerprint classifier fusion with update rule to minimize training time

This paper presents a novel fingerprint classifier fusion algorithm using Dempster-Shafer theory concomitant with update rule. The proposed algorithm accurately matches fingerprint evidences and also efficiently adapts to dynamically evolving database size without compromising accuracy or speed. We experimentally validate our approach using three fingerprint recognition algorithms based on minutiae, ridges, and image pattern features. The performance of our proposed algorithm is compared with these individual fingerprint algorithms and commonly used fusion algorithms. In all cases, the proposed Dempster Shafer theory with update rule outperforms existing algorithms even with partial fingerprint image. We also show that as the database size increases, the proposed algorithm is designed to operate on only the augmented data instead of the entire database, thereby reducing the training time without compromising the verification accuracy.