Quantitative Evaluation of Normalization Techniques of Matching Scores in Multimodal Biometric Systems

This paper attempts to make an quantitative evaluation of available normalization techniques of matching scores in multimodal biometric systems. Two new normalization techniques Four Segments Piecewise Linear (FSPL) and Linear Tanh Linear (LTL) have been proposed in this paper. FSPL normalization techniques divides the region of genuine and impostor scores into four segments and maps each segment using piecewise linear function while LTL normalization techniques maps the non-overlap region of genuine and impostor score distributions to a constant function and overlap region using tanh estimator. The effectiveness of each technique is shown using EER and ROC curves on IITK database of having more than 600 people on following characteristics: face, fingerprint, and offline-signature. The proposed normalization techniques perform better and particularly, LTL normalization is efficient and robust.