Deep Learning Driven Palmprint Recognition Using Smartphone-Based Video Dataset

In recent years, biometric systems have become increasingly prevalent, offering a reliable method for verifying individuals’ identities. Among the various biometric identifiers, palmprint recognition has emerged as a particularly promising option due to its high user acceptability and the uniqueness of palm features. This paper presents a comprehensive study on palmprint recognition using deep learning models, with a specific focus on utilizing a smartphone-based video dataset. The study involves the collection of a diverse video dataset using smartphone cameras, which provides a rich set of palmprint images under various palm position and lighting condition. From this video dataset, individual frames are extracted and processed to create a standardized image dataset personalized for deep learning applications. This dataset is then employed to train and evaluate six different deep learning architectures: DenseNet 121, DenseNet 201, Xception, Inceptionv3, MobileNetv2, and NASNet. Among the models tested, DenseNet 201 emerges as the most accurate one.