Development of a real-time face recognition system using deep learning techniques

Face recognition system is a non-invasive type of biometric system used to verify or identify a person from a digital image. Face recognition systems are mainly used for security purposes; however, they can be used in many other areas such as entertainment and healthcare. A lot of work has been done in the literature about the accuracy and speed of face recognition systems over the years. In this thesis, a real-time face recognition system has been designed and implemented on 2 different inexpensive platforms. The system used a combination of different machine learning and deep learning algorithms and techniques. The face recognition system has been implemented in four main stages. For the face detection, the Histogram of Oriented Gradients (HOG) was used because it is faster in detecting faces on a digital image. After detecting the face a modified version of face landmark estimation algorithm was used to generate 5 face landmarks used to center the face before passing it on to a pre-trained face recognition model which generates 128 embedding for the face. Finally, the system used a Support Vector Machine (SVM) classifier to identify whose face is on the image. The system reached a performance of 96.88% accuracy when tested with a database of 40 images of 8 different individuals. Approximately 3 frames have tested per second. This thesis shows that real-time face recognition systems based on recent deep learning techniques can be implemented on a limited computer hardware. Key Words: Face recognition, deep learning, machine learning, computer vision