A Hybrid LBP-CNN Feature Fusion Framework Using SVM for Robust Face Anti-Spoofing Attack Detection

Technologies working behind recognition and biometric authentication systems are increasing day by day so as the attacks posed by them. The security is a major concern that makes these systems vulnerable to presentation or anti spoofing attacks like 2D attacks such as printed photo attacks. Existing Face Anti Spoofing research has improved detection capability, but practical deployment still faces challenge such as obtaining sufficiently rich discriminative representations, controlling computational complexity, avoiding dependence on specialized sensing hardware, and maintaining performance when the data distribution changes. To counter these challenges this paper proposes a lightweight monocular RGB anti spoofing attack detection framework that can extract complementary features from a single conventional camera. Multi-Scale Local Binary Pattern provides multi-scale local texture information hierarchical fusing Convolutional Neural Network providing hierarchical learned semantic features and Self-Attention emphasizing on informative spatial regions. Global Average Pooling providing deep semantic information. Classification performed using a hyperparameter-optimized Support Vector Machine (SVM). The monocular architecture enables feature extraction, feature fusion, and classification components to be independently enhanced while maintaining computational efficiency. Experimental evaluation demonstrates an overall accuracy of 97.26\%, with Precision of 99.21\%, Recall of 93.52\%, Specificity of 99.57\%, F1-score of 96.28\%, and Balanced Accuracy of 96.54\%. The architecture also attains an MCC of 0.9431 and Cohen's Kappa of 0.9421. Cross-dataset evaluation obtains approximately 93\% accuracy, demonstrating promising generalization to an independent dataset. The results indicate that combining complementary local texture and hierarchical deep representations provides an effective and computationally efficient solution for RGB-based detection of planar presentation attacks.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-19
DOI
https://doi.org/10.21203/rs.3.rs-11511764/v1
Primary Topic
Biometric Identification and Security
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

A Hybrid LBP-CNN Feature Fusion Framework Using SVM for Robust Face Anti-Spoofing Attack Detection

Yudhishthir Raut, Deepak Kumar Jain, Naimisha Awasthi, Usha Sharma
Research Square
Biometric Identification and Security
preprint

A Hybrid LBP-CNN Feature Fusion Framework Using SVM for Robust Face Anti-Spoofing Attack Detection

Yudhishthir Raut, Deepak Kumar Jain, Naimisha Awasthi, Usha Sharma
preprint en

Abstract

Technologies working behind recognition and biometric authentication systems are increasing day by day so as the attacks posed by them. The security is a major concern that makes these systems vulnerable to presentation or anti spoofing attacks like 2D attacks such as printed photo attacks. Existing Face Anti Spoofing research has improved detection capability, but practical deployment still faces challenge such as obtaining sufficiently rich discriminative representations, controlling computational complexity, avoiding dependence on specialized sensing hardware, and maintaining performance when the data distribution changes. To counter these challenges this paper proposes a lightweight monocular RGB anti spoofing attack detection framework that can extract complementary features from a single conventional camera. Multi-Scale Local Binary Pattern provides multi-scale local texture information hierarchical fusing Convolutional Neural Network providing hierarchical learned semantic features and Self-Attention emphasizing on informative spatial regions. Global Average Pooling providing deep semantic information. Classification performed using a hyperparameter-optimized Support Vector Machine (SVM). The monocular architecture enables feature extraction, feature fusion, and classification components to be independently enhanced while maintaining computational efficiency. Experimental evaluation demonstrates an overall accuracy of 97.26\%, with Precision of 99.21\%, Recall of 93.52\%, Specificity of 99.57\%, F1-score of 96.28\%, and Balanced Accuracy of 96.54\%. The architecture also attains an MCC of 0.9431 and Cohen's Kappa of 0.9421. Cross-dataset evaluation obtains approximately 93\% accuracy, demonstrating promising generalization to an independent dataset. The results indicate that combining complementary local texture and hierarchical deep representations provides an effective and computationally efficient solution for RGB-based detection of planar presentation attacks.

Research Square
Dr. A.P.J. Abdul Kalam Technical University (IN), Symbiosis International University (IN), Dalian University of Technology (CN)
Biometric Identification and Security
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon