.png){kind=icon}
.png){kind=icon}
Return
Privacy-enhanced multimodal biometric authentication via deep binarization and feature-hashed projection
.png){kind=icon}
PRE
AI
R
S
K
D
DOI：10.1117/1.JEI.35.2.023024.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Although compared with unimodal systems, multimodal biometrics are more reliable and spoofing resistant, which has led to their growing popularity. Among various biometric traits, iris and fingerprint modalities are widely preferred for their high distinctiveness and ease of acquisition. However, protecting the stored biometric templates remains a critical concern to preserve user privacy. We propose a cancelable multimodal biometric template protection scheme based on deep binarization and feature-hashed random projection (FHRP) to secure iris and fingerprint templates. Initially, discriminative feature extraction was performed for both modalities using a pre-trained Convolutional Neural Network (CNN). These features are then binarized through a deep binarization process and fused using an addition-rotation-XOR (ARX) operation to obtain a robust feature-representation. Subsequently, the fused features are transformed using FHRP to produce a secure and non-invertible template, which is stored on the server for authentication. Experimental evaluations on three publicly available datasets-Children Multimodal Biometric Database as Database-1, CASIA-V1 & FVC-2004 as Database-2, and CASIA-V3 & FVC-2006 as Database-3-demonstrate that the proposed approach achieves superior recognition accuracy and lower equal error rate compared with existing techniques and effectively safeguarding user privacy.
Keywords:
template protection
multimodal biometrics
deep binarization
neural networks
security
Journal
J
IF:
1
Papers:
109
Citations:
2.7K
Organization
Cited Papers