Efficient palm vein biometric authentication system using hybrid PCANet deep learning and hyper parameter machine learning

Palm vein authentication uses an individual’s palm’s intricate vascular patterns as a unique identifier. The veins and other subdermal features, which are substantial, resilient, stable, and obscured mainly beneath the skin, are inherently problematic to counterfeit. Despite extensive research in this domain, a persistent problem remains in the suboptimal quality of data collection and processing. This lack of high-quality data has led to substandard authentication performance and failures in the feature extraction process. The presented study suggested an efficient Palm Vein biometric authentication system (EPV-BAS) depending on a hybrid PCANet Deep Learning (DL) and Hyper Parameter Machine Learning (HPML) to address these issues. Two ML models are integrated into the well-known hybrid PCANet DL framework: a straightforward search for Model Parameters as well as a random search for the HyperParameters. It applies Multi-Layer Perceptron (MLP), Support Vector Machine (SVM), and K-Nearest Neighbors (KNN) techniques. Additionally, the suggested EPV-BAS introduces three image fusion methods—Gabor filter (GF) images, high-frequency emphasis filtering (HFEF), and contrast-limited adaptive histogram equalization (CLAHE)—that together provide an advanced strategy for palm vein image augmentation. In terms of the PLUSVein-Contactless hand veins at 850 nm and 950 nm, the suggested EPV-BAS worked better. Every dataset was split up into sets for training and testing. With an F1-score in certain metrics (98.25% and 96.86%) for KNN, (99.57% and 99.42%) for SVM, and (99.77% and 99.34%) for MLP, PLUSVein-Contactless hand vein RL850nm dataset. The PLUSVein-Contactless hand vein RL950nm dataset has several metrics with an F1-score: (100% and 99.13%) for MLP, (98.24% and 97.21%) for KNN, and (99.54% and 97.15%) for SVM. These outcomes cover a random search model without and with PCA dimensionality reduction for all three approaches. In terms of performance, the EPV-BAS outperforms the most advanced palm vein authentication techniques.