CSFN: Cancelable SoftmaxOut Fusion Network based Multimodal Biometric Template Security

Biometric applications are widely used for personal verification for the identification of individuals; however, existing issues include the security of the template, the identification accuracy, and computational complexity. To overcome these drawbacks, the present Cancelable SoftmaxOut Fusion Network (CSFN) solves them by establishing a fusion of iris and face data modalities to acquire safer cancelable templates. The Cancelable SoftmaxOut Fusion Network (CSFN) eliminates the need for traditional methods of feature fusion and transformation, resulting in its effectiveness and high level of security. Multimodal feature extraction is done, and probability outputs are obtained via the SoftmaxOut function to fuse the features and user-specific transformations while also performing distance-preserving compression and cryptographic hashing to produce the CSFN templates. The system was compared with two models, Multifeature Deep Learning Network (MDLN) and the Hierarchical Fusion Network (HFN) on the CASIA-Iris, IITD Iris, and LFW (Labeled Faces in the Wild) databases for face recognition. The results show that the proposed CSFN yields superior performance compared to other methods, with the least EER of 1.2% and the maximum GAR of 97.8% for all datasets. Its implementation time, with an average template generation time of 8.5 ms and an average template verification time of 3.2 ms, also demonstrates its feasibility. Nevertheless, CSFN proves better with low FAR and FRR values, indicating better recognition capability and system security. Still, CSFN guarantees that it can't be linked, reversed, or revoked. This makes it the better or more efficient way to use scale-based multimodal biometric authentication today.