Publication

Abstract : Unimodal biometrics has various issues: noisy data, inter-class similarities, intra-class similarities, spoofing, and non-universality, making this system less secure and accurate. Compared to this Unimodal, multi-modal biometrics has various benefits such as high accuracy, low error rate and larger population coverage. Generally, multi-modal biometrics systems enhance integrity and privacy, as it stores different biometric characteristics of every user. Therefore, the multi-modal biometric template must be protected to avoid leaking biometric information to an adversary. An innovative multi-modal system is introduced, combining two biometric traits, fingerprint and retina. However, the deep learning (DL) approach will be hybrid with the hashing method to improve the template’s security. In recent years, DL has presented a great demand in multi-modal biometric devices for feature extraction. This proposed methodology consists of pre-processing, feature extraction, and hashing. Pre-processing takes place for image resizing and augmentation purposes. In the feature extraction stage, CNN has been used to extract spatial information. It doesn’t provide a correlation between the adjacent features. To solve this, efficiently combine the CNN and BiGRU model with the self-awareness layer (ConvGRUSA). In ConvGRUSA, the parameters are tuned using the tunicate swarm optimization (Tuni-SO) algorithm. With the aid of BiGRU, the correlation between the adjacent features is detected. Finally, the self-attention layer fuses the deep features using spatial-wise weighting and channel-wise weighting. After extracting the deep features, the chaos-based hash function will be proposed to generate a secure multi-modal template. The work will be implemented on the MATLAB platform, and the performance metrics of accuracy, precision, FPR, FNR, FAR, FRR, TPR, TNR and EER will be evaluated. The proposed method achieves an accuracy performance of 99.93, FAR is 0.11, FRR is 0.18, and EER is 0.15. The proposed method is compared with some existing methods to prove the effectiveness of the proposed system.