A Metaheuristic-Optimized Deep Neural Network for Feature-Level Fusion in Multimodal Biometrics for Secure Healthcare Systems

Abstract

Biometric security in healthcare systems is quickly becoming an important issue in the data security industry. An enormous obstacle remains the development of multimodal biometrics systems for smart environments that improve detection rates and accuracy. Traditional approaches using varying degrees of traits fusion are contrasted with MBS that use Convolutional Neural Networks (CNNs). This paper details the creation of a multimodal biometric identification system that enhances the accuracy and security of the system by fusing fingerprint and iris scanning at the feature level in smart healthcare systems. For feature extraction, it makes use of the Gray-Level Co-Occurrence Matrix (GLCM), Histogram of Oriented Gradients (HOG), and Local Binary Pattern (LBP). The Chimp Optimization Algorithm (ChOA) feature optimizer and Grey Wolf Optimizer (GWO) are utilized for feature selection. Using a Resnet, Densenet, and suggested Deep Neural Network (DNN) is the last stage in the categorization process. The results indicate that the DNN with ChOA outperformed DNN with GWO optimizer, with an accuracy rate of 98.14 %, AUC OF 0.992, F1-Score of 91.2 % and kappa with 0.959. This evidence is believed to using DNN with ChOA optimizer is an optimum choice for multimodal biometric system in smart healthcare applications.