Face Recognition System using Dual Network

Face recognition systems are the most widely used biometric technologies due to three reasons: collectability, distinctness and the rise of artificial intelligence. Deep learning-based face recognition systems are robust but require a balance between performance and speed. Additionally, recognition accuracy can be severely impacted by unconstrained environments. This paper proposes a dual network face recognition system for the two main tasks, which are face detection and face recognition. E-Face was introduced for face detection, while FRM is presented as a solution for face recognition. The E-Face uses EfficientDet architecture and has achieved good accuracy of 91.5% in the UTKFace dataset. Meanwhile, FRM uses a depthwise separable convolution network as the base of its architecture and three key designed elements. This help improve its feature extraction capabilities without increasing the size and complexity of the model. The design elements are the interval usage of dilation convolution, squeeze-and-excitation modules and spatial attention modules. FRM has achieved a great face recognition accuracy of 96% in the filtered LFW dataset.