Real-time Face Identification from Video in an Uncontrolled Environment using CNN

We have witnessed a significant amount of fraud and security issues in modern life. Numerous biometric characteristics, such as the eyes, face, fingers, and palms, are used to address these problems. Among these, facial recognition is considered one of the least intrusive methods and is frequently used to identify or verify an individual. Face recognition is one of the most effective applications of computer vision, and has achieved considerable attention in recent years. Deep learning networks have achieved state-of-the-art performance in still-image-based face recognition. Video-based face recognition is a more complex task than still-image-based face recognition due to video quality, pose variation, occlusion, and illumination, and it also entails processing a large volume of data. We address these challenges by developing an efficient deep learning model trained, tested, and evaluated on the YouTube Face Dataset, designed for unconstrained face recognition in videos. In this paper, a deep learning face detection algorithm, Multi-task Cascaded Convolutional Neural Network (MTCNN), is employed to detect and localise faces in videos. Feature extraction and face recognition have been performed by using a convolutional neural network (CNN). This model has been proposed for accurate face detection and recognition from unconstrained video and performs better on the YouTube face dataset. The test accuracy of the proposed model is 93.11%. This work has been conducted to improve face recognition accuracy in the presence of intra-video variations.