Integrated Single Shot Multi-Box Detector and Efficient Pre-Trained Deep Convolutional Neural Network for Partially Occluded Face Recognition System

Partial occlusion is a key issue in face recognition as it decreases the recognition accuracy. As a result, current face recognition systems are limited to operate under constrained environments. To resolve the partial occlusion problem, we propose a system that adopts the convolutional neural network but with a pre-trained model for robust face recognition and facial feature extraction. The model improves the accuracy of partially occluded face recognition. Moreover, the face detection network utilizes the feature pyramids to reduce the number of network parameters and achieve scale invariance. The image context module is also incorporated to increase the receptive field, as it effectively improves the detection accuracy and reduces memory usage. Experiment results show that the proposed method performs better than existing state-of-the-art methods for the detection and recognition of occluded faces.