Citation

Ali, Md Farman and Noor, Jahid Hassan and Nabi, Md Serajun and Abdul Karim, Hezerul and Samia, Sadia Afrin and Mahmud, S M Nadim (2025) Hybrid Feature-Based Machine Learning for Real-Time Biometric Facial Liveness Detection. In: 2025 Multimedia University Engineering Conference, MECON 2025, 21 July 2025 - 23 July 2025, Cyberjaya, Malaysia.

Text 8.pdf - Published Version Restricted to Repository staff only Download (1MB)

Abstract

Face recognition is now the most used biometric mode; however, it is vulnerable to spoofing. Spoofing attacks include image-printed attacks, video replay attacks, 3D mask attacks, and so on, necessitating the implementation of appropriate anti-spoofing methods. Previous methods for early liveness identification used handcrafted techniques such as Local Binary Patterns (LBP) or Histogram of Orientated Gradients (HOG) to analyze texture variations between real and synthetic faces. Without a doubt, these methods were computationally efficient, but they proved ineffective against sophisticated attacks such as high-resolution printouts or 3D masks. Aside from the methodologies used in the referred models, researchers interested in improving face identification presented CNN (Convolutional Neural Networks) solutions. However, despite their high accuracy in terms of outcomes, deep-learning methods are computationally expensive, require large amounts of data, and are not generalizable when compared to other spoofing approaches. As a result, the paper suggests a lightweight strategy that does not use deep learning but integrates security, efficiency, and scalability into the output. It distinguishes between authentic, and spoof faces by utilizing handmade extraction techniques such as histograms of orientated gradients (HOG), local binary patterns (LBP), and Gabor. These collected characteristics are identified using machine learning-optimized ensemble models such as Support Vector Machines (SVM), Random Forest (RF), and XGBoost, which ensure high generalization while requiring minimal CPU resources. This model is up to 99% efficient in terms of accuracy and processing resources. This is one of the first experiments to achieve more than 99% accuracy in facial recognition using a lightweight, non-deep learning technique. Unlike prior research, which concentrated on CNNs or specialized customized techniques, our hybrid system combines LBP, HOG, and Gabor features with ensemble machine learning classifiers (SVM, RF, XGBoost) to provide real-time, scalable security suitable for low-resource applications. Keywords: Facial Liveness Detection, Biometric Security, Anti-spoofing, Digital Identity.

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	Biometrics, real-time face recognition
Subjects:	T Technology > TA Engineering (General). Civil engineering (General) > TA1501-1820 Applied optics. Photonics
Divisions:	Faculty of Artificial Intelligence & Engineering (FAIE)
Depositing User:	Ms Rosnani Abd Wahab
Date Deposited:	18 Mar 2026 08:20
Last Modified:	19 Mar 2026 02:12
URII:	http://shdl.mmu.edu.my/id/eprint/15588

Downloads

Edit (login required)