Video Spoofing Attack Detection Using Convolutional Neural Networks and Ensemble Learning Voting Classifier Methods

Facial recognition technology is increasingly being used in various applications, but this has resulted in the emergence of new threats such as spoofing. Existing detection systems still face several shortcomings, such as low accuracy or limited variety of training data, making them vulnerable to spoofing attacks. This research develops a spoofing detection system based on Convolutional Neural Networks (CNN) and ensemble learning methods that combine several models, such as Support Vector Machines (SVM), Random Forests, and Logistic Regression with Voting Classifier techniques tested on the iBeta Level 1 - Liveness Detection Dataset. This approach is done by utilizing a combination of models to improve detection accuracy and reduce the weakness of individual models. The proposed system is tested using validation and test datasets to ensure no overfitting/underfitting occurs. The experimental results in the test data in this study show that the method achieves 97% accuracy on tests and 98% on validation for ensemble learning, as well as 100% on test and validation for CNN-based models. These findings prove the effectiveness of deep learning and ensemble learning approaches in detecting spoofing, thus potentially improving the security of face recognition systems.