Rank level fusion in multibiometric systems

Abstract

Multibiometric systems have recently become a preferred option for human identification over the unibiometric systems. It increases the recognition rate and confidence in the final decision, and simultaneously reduces the failure to enroll rate (FER). For identification mode, rank level fusion is a feasible option as incompatibility and normalization issues present at the score level fusion are not prominent at this level and also sufficient information is present to fuse as opposed to the decision level fusion. We propose an improvement in existing rank level fusion techniques using two levels of hierarchy. Series and parallel combinations are proposed to combine the output of various rank level fusion techniques. Two formulations of series and parallel combinations are extensively evaluated on multi-algorithm, multiinstance and multi-modal biometric systems created from three publicly available datasets: (i) NIST BSSR1 [1] multimodal biometric score database, (ii) Face Recognition Grand Challenge V2.0 [2] and (iii) LG4000 [3] iris images.

A Comparative Analysis of Decision-Level Fusion for Multimodal Driver Behaviour Understanding

“…A theoretical study of such methods from the pre-Deep Learning era is provided in [26]. Rank-level decisionlevel fusion, such as Borda Count voting [27], [28], [29] and Reciprocal Rank Voting [30] are less popular, but have been successfully applied in the field of biometric identification [31], [32]. There are several works targeting multimodal fusion through learning-based methods, e.g., using SVM, LSTM, or neural network fusion layers [33], [34], [35], [7], [36], [37].…”