Stock exchange regime algorithm optimized deep learning framework for hybrid multimodal biometric authentication

Abstract

Biometric recognition is crucial for secure and reliable access control in high-security systems that include surveillance, law enforcement, and smart cities. Though the Deep Learning (DL) models present exceptional performance in biometric recognition, a few challenges faced by biometric recognition are revealing an individual’s identity when stolen and falsifying legal documents, passports or criminal records. To alleviate these kinds of issues, this work proposes a new multimodal biometric authentication model for providing recognition using fingerprint and finger vein images. Initially, both images undergo pre-processing by an Adaptive Weiner Filter to discard the noise from the images. Later, the minutiae points are detected by Hit-or-Miss Transformation, and then the authentication phase is performed by SqueezeNet. The SqueezeNet is trained with the Stock Exchange Regime Algorithm (SERA). Similarly, the finger vein images are forwarded to the pre-processing phase to remove the noise, which is then followed by vein extraction that is implemented using Repeated Line Tracking. Here, the output attained from the vein extraction phase is given to the feature extraction phase that uses grid-based location to form features, and then it is forwarded to the authentication process, which is done by a Deep Neuro Fuzzy Network (DNFN). Finally, after the authentication process, the outcomes are being trained by the proposed Stock Exchange Regime Algorithm (SERA), integrating both Stock Exchange Trading Optimization (SETO) and Flow Regime Algorithm (FRA). The outputs obtained from SqueezeNet and DNFN are combined by Motyka similarity to determine the outcome. The SERA-based SqueezeNet+DNFN model demonstrates strong performance across multiple evaluation metrics, achieving an accuracy of 94.5%, sensitivity of 91.3%, specificity of 95.1%, an Equal Error Rate (EER) of 0.078, a False Acceptance Rate (FAR) of 0.089, and a False Rejection Rate (FRR) of 0.083. These results correspond specifically to the combined fingerprint and finger vein modality using dataset-1.