Matching Score-Based Biometric Quality Estimation for Score and Rank Level Fusion for Person Identification

Abstract

A unimodal biometric system is often affected by inter-class similarity, intra-class variation, non-universality, and poor data quality. A multimodal biometric system combines multiple biometric traits to overcome these problems. Moreover, multimodal biometric system often achieves better recognition performance. Nevertheless, several existing fusion approaches derelict the impact of the quality of the given biometric signal while fusing. It is essential to incorporate the quality-related information of a given biometric signal into the recognition system to demean the effect of a biometric signal with poor quality. In this work, a novel biometric quality estimation approach is proposed. The proposed approach deduces the quality information from the matching scores between a probe and a gallery in an identification setup. In order to assess the quality of the biometric signal, the proposed approach estimates the distinctiveness of the best matching score with respect to few next best matching scores. The estimated quality is used as weight for individual modalities while fusing the information. Experiments involving state-of-the-art score level and rank level fusion methods establish the superiority of the proposed quality estimation approach over existing biometric quality estimation approaches.