An Efficient Palmprint Authentication System based on One-Class SVM and HOG Descriptor

Biometrics characteristics are usually employed in the context of authentication or identification. They are particularly relevant when designing security systems aiming to protect private data and ensure a certain level of access security. Several physiological and behavioural characteristics might be used depending on the type of available sensors. In the context of this paper, we propose a biometric-based authentication system using palmprint as a modality. We introduce the use of one-class classification using a well-known machine learning technique namely the Support Vector Machine classifier during the authentication process. Its main advantage lies in increasing the computational efficiency since each classifier is dedicated to each specific user which makes it independent of the others and reduce the computational load. Moreover, we propose to employ the Histogram of Oriented Gradients along with Principal Component Analysis to generate a relevant and discriminant representation of the palmprint image. We evaluated the proposed biometric-based authentication system with a public benchmark dataset and obtained state-of-the-art performance with 94.67% accuracy.