FACE LIVENESS DETECTION USING CNNALGORITHM(DL)

Abstract

Face recognition has become a widely adopted biometric authentication method due to its convenience, non-intrusiveness, and ability to authenticate users in realtime. However, the widespread adoption of face recognition systems has also led to a surge in spoofing attacks, where attackers attempt to bypass security by using printed photographs, digital images, replayed videos, or even 3D masks. Traditional face recognition algorithms are unable to differentiate between genuine live faces and these spoofing attempts, which poses a critical risk to security systems in banking, mobile authentication, border security, and other sensitive applications. Therefore, ensuring that the detected face is live, a process known as face liveness detection, has become a crucial requirement for secure biometric authentication. Recent advancements in deep learning have provided powerful tools for solving complex image-based recognition tasks. Among these, Convolutional Neural Networks (CNNs) have demonstrated remarkable performance in image classification, object detection, and feature extraction tasks due to their ability to learn hierarchical features from raw images without manual feature engineering. This paper proposes a CNN-based approach for face liveness detection, designed to distinguish between live and spoofed faces with high accuracy and robustness under varied conditions, including different lighting environments, backgrounds, and facial orientations. The proposed system leverages CNN’s hierarchical feature extraction to capture subtle differences between real and fake faces that are often imperceptible to human observation, such as texture inconsistencies, reflection patterns, and pixel-level variations. The proposed method was evaluated using widely recognized datasets, including CASIA-FASD and Replay-Attack, which provide diverse scenarios and various types of spoofing attacks. Preprocessing steps include face detection, resizing, and normalization, along with data augmentation techniques such as rotations, flips, and brightness adjustments to improve the model’s generalization. The CNN architecture consists of multiple convolutional layers for feature extraction, pooling layers for dimensionality reduction, and fully connected layers for classification. The model is trained using the Adam optimizer with categorical crossentropy loss, and its performance is evaluated in terms of accuracy, precision, recall, and F1-score.