A comparative study between handcrafted features and deep features for biometric systems

Abstract

Biometrics involves identifying individuals based on their physiological or behavioral traits. Biometric systems are essential tools in many security-related applications. This dissertation compares two feature extraction approaches : handcrafted methods such as LPQ and ML-LPQ and deep learning-based methods using neural networks like AlexNet and DenseNet-201. Palmprint images were used as the biometric data source. The eva- luation focused on accuracy, processing speed, and computational complexity. The results show that deep features offer higher accuracy, while handcrafted methods are simpler and faster. Combining both approaches can lead to more balanced and effective biometric systems.