A Deep Learning Approach for Recognition Systems

Biometrics is emerging as one of the most reliable solutions for identifying individuals in modern security systems. Unlike passwords, it relies on physiological characteristics unique to each person, offering a higher level of security.This book focuses on two promising modalities: finger veins and palm prints. While stable and difficult to forge, they remain sensitive to variations in lighting and the quality of the captured images. An approach combining deep learning and machine learning is proposed. Features are extracted using pre-trained convolutional neural networks (VGG16, VGG19, MobileNetV2) and then refined through \textit{fine-tuning}. SVM, KNN, and Random Forest classifiers are then applied, with a comparison between single-instance and multi-instance approaches. MobileNetV2 offers the best performance in terms of accuracy and efficiency. The multi-instance approach enhances the system's robustness, while SVM stands out for its recognition accuracy.