Decision level fusion schemes for a Multimodal Biometric System using local and global wavelet features

Abstract

Local and global features of biometric data play a vital role to improve the performance of a multimodal biometric system. In this paper, we propose three decision level fusion schemes - Local Decision Fusion (LDF), Global Decision Fusion (GDF) and Local-Global Decision Fusion (LGDF)- by exploiting local and global information. The proposed methods extract such information by utilizing low and high frequency wavelet sub-bands. Subsequently, the sub-bands are classified separately using nearest neighbor classifier, and the resulting classes are fused using weighted majority voting. The proposed LDF and GDF methods show an improvement in the average recognition rates to a maximum extent of 9.4%, 11%, 10.6% and 11.5% in comparison to unimodal LDF, GDF and two low frequency sub-band based methods respectively. Further, the proposed LGDF method is superior to feature-score hybrid fusion by a highest average recognition rate of 6.75%.