Analysis of Deep Learning Frameworks for Occluded Face Recognition

Face recognition has been fully applied in more and more fields, such as security monitoring, financial services, social media and traffic management. However, masks, glasses, and other items often block the face, which is also affected by changes in light and posture. Therefore, efficiently dealing with face recognition based on occlusion is very important. With the development of deep learning, many neural network frameworks have been effectively applied to this problem. This paper summarises the process of occluded face recognition (OFR) and introduces some standard face recognition datasets. Convolutional Neural Networks (CNN), Variational Autoencoder (VAE) and Generative Adversarial Networks (GAN), which are three deep learning frameworks used for OFR, are demonstrated, and the last two of them are compared. A hybrid model of VAE and GAN is proposed. This model can avoid the common pattern breakdown problem of GAN and improve the stability of the training process. At the same time, the model can generate multiple possible occlusion cases and more realistic images, effectively improving the ability to process occlusion.