الفهرس 
Chapter 1. INTRODUCTIONOnly 14 pages are availabe for public view
 |  | from | 66 |  |  |
| | | from | 66 | | |
Abstract
As we step further into the digital age, facial recognition technology is becoming ubiquitous, deeply integrated into areas such as personal device security, financial services authentication, and even public safety. However, with this technological advancement comes a pressing challenge: the risk of spoofing attacks. These attacks, where a fraudulent user attempts to gain unauthorized access, present a significant threat to the integrity of facial recognition systems. This brings us to the critical importance of developing a face liveness detection system that can distinguish between a real human face and a fake one. It is an essential layer of defense that ensures the person in front of the camera is physically present and not a photo, video, mask, or other forms of spoof.
In this thesis, we propose two novel approaches to enhance face liveness detection and improve the authentication process. The first proposed solution, ”Restrictive Voting Technique for Faces Spoofing Attack”, leverages an ensemble of multiple classifiers with specialized features for spoofing attack detection. By employing a restrictive voting scheme, our approach effectively combines the strengths of individual classifiers, resulting in superior performance compared to state-of-the-art methods. Through comprehensive evaluations and analysis, our proposed technique demonstrates its effectiveness in detecting face spoofing attacks. While traditional evaluation metrics such as Equal Error Rate (EER) and Half Total Error Rate (HTER) provide a balanced assessment of the system’s accuracy for both legitimate users and impostors, our analysis suggests that they may not be the most appropriate metrics for evaluating the effectiveness of a face spoofing attack detection system. Instead, we prioritize the reduction of False Acceptance Rate (FAR) to prevent unauthorized access by impostors, while aiming for a small HTER. Our technique shows promising results in reducing the FAR and maintaining a low HTER, highlighting its capability in accurately distinguishing between genuine and spoofed faces.
The second proposed solution, ”Transfer Learning Approach for Face Liveness Detection”, harnesses the power of transfer learning and pre-trained deep neural network models. By fine-tuning a pre-trained model on a moderate-scale dataset, our approach captures discriminative features specific to liveness detection. It outperforms existing methods in terms of accuracy, precision, recall, and F1-score, and demonstrates superior performance in distinguishing between genuine and spoofed faces. Comparative analysis against state-of-the-art techniques confirms the effectiveness of our approach in real-world scenarios.
Through comprehensive evaluations, both proposed solutions contribute to enhancing face liveness detection. The Restrictive Voting Technique harnesses the power of ensemble learning, while the Transfer Learning Approach benefits from knowledge learned from large-scale datasets. These findings establish a strong foundation for more secure and reliable face recognition systems, adding noticeable contributions to the field of face liveness detection, providing insights and methodologies for improving biometric authentication systems. The proposed solutions address inherent challenges and lay the groundwork for future advancements in securing facial recognition technology.