Abstract :

Single-modality Biometric Authentication, Relying On One Biometric Trait Matched By A Classical Distance-based Algorithm, Remains Attractive For Its Simplicity But Is Fundamentally Limited In Forensic Settings Where Degraded, Rotated, Or Partially Occluded Samples Are The Norm Rather Than The Exception. This Paper Presents A Forensic Multimodal Biometric Authentication System That Fuses Fingerprint, Face, Iris, And Voice Evidence Through A Configurable Feature-level, Score-level, Or Decision-level Fusion Engine, With The Fused Matching And Accept/reject Decision Realized In A Dedicated FPGA Hardware Core Alongside The Software Machine-learning Pipeline. The Existing Single-modality Fingerprint Baseline, Evaluated On Real FVC2004 Set B Data With A Nearest-neighbor Minutiae Matcher, Achieves 35.4% Accuracy, A Perfect 0.0% False Acceptance Rate, And A High 73.8% False Rejection Rate. The Proposed Systems Learned SVM Fingerprint-branch Classifier, Trained And Tested On The Identical Dataset And Split, Raises Accuracy To 58.9% And Reduces The Equal Error Rate From 35.7% To 41.1% While Substantially Rebalancing Acceptance And Rejection Behavior, With A Real, Better-thanchance ROC AUC Of 0.645. On The Hardware Side, The Fusion And Decision Core Synthesized For A Xilinx Artix-7 Device Occupies Only 261 Post-route Slice Look-up Tables And 396 Registers, Draws 0.104 W Of Total On-chip Power, And Closes Timing At 100 MHz With A Positive Worst-negativeslack Margin Of 2.648 Ns. These Results Demonstrate That A Learned, Fusion-capable Authentication Pipeline Realized Through A Lightweight, Timing-closed, Sub-watt FPGA Core Is A Practical And Forensically Motivated Upgrade Path Beyond Single-modality Software-only Matching.

Published:

27-7-2026

Issue:

Vol. 26 No. 7 (2026)

Page Nos:

1328-1336

Section:

Articles

License:

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

How to Cite