Bring your own hand: how a single sensor is bringing multiple biometrics together

Abstract

In spite of innumerable benefits of multi-biometry authentication, a couple of imaging sensors are required to capture multi-biometric samples, but that increase the overall cost as well as degree of user cooperation. This work presents a single sensor-based multimodal biometric identification system by fusing major finger knuckle, minor finger knuckle, palm print and handprint features of the human hand for enhancing the security and privacy of any consumer device. A virtual imaging device has been suggested to capture palmer and dorsal view of hand with single-shot multi-trait acquisition mechanism. The hand images captured from digital camera are first preprocessed to get the major knuckle, minor knuckle and palm ROI’s. The finger knuckles and palm print ROI’s are then enhanced and transformed to illumination invariant representation using robust encoding techniques, over which ray tracing features are emphasized predominantly. A non-rigid multi-scale approach, namely deep matching, has been employed to obtain the matching score between the corresponding correlation maps for finger knuckle or palm print recognition. Apart from that, we present a new scheme to extract shape, and geometrical features of handprint and employ metric learning-based L2-norm for feature matching in $$2D^{2}{} \textit{PCA}+2D^{2}{} \textit{LDA}$$ space. Five publicly available databases are used to evaluate the effectiveness of proposed approach. Finally, the score-level weighted sum rule fusion has been adopted to combine matching scores of four traits which show that the proposed method outperforms other unimodal and state-of-the-art multimodal identification methods in terms of EER (0.01%), DI (3.64) and CRR (100%).