Literature Survey on Face Recognition of Occluded Faces

In this paper, a study of the recent advancements in occluded face recognition is made. Recently, this has become very relevant due to the outburst of corona virus. Because of the seriousness of the spread of the disease, every person wore a mask, and it became challenging for automated systems to identify the person. So, it became the need of the hour to find new techniques for occluded face detection. In the years 2021 to 2022, many papers have been published in this regard. Researchers all over the world have contributed significantly towards occluded face recognition. Also, the advancement of deep learning has helped researchers a lot in implementing innovative techniques to deal with the challenges in face recognition, especially occluded faces, which is considered one of the most difficult challenges to overcome. This paper intends to push the limits further and encourage researchers to find more innovative techniques for occlusion-independent face recognition. In this paper, the recent publications in this field are studied. This paper gives an insight into different algorithms which use Convolutional Neural Networks, Deep Learning, Attention mechanism, Dictionary representation, Simultaneous segmentation, Joint segmentation and identification, etc., for occluded face detection and recognition. Incorporating these strategies in face recognition has substantially improved the recognition rate, specifically for occluded face recognition. Additionally, the databases for training occluded face recognition algorithms lack adequate data. This paper also gives an insight into the recent databases available for training occluded face recognition algorithms. Also, different methods for obtaining synthesised occluded faces are discussed.

I. Introduction

Face recognition has always been an interesting topic for researchers because of its relevance in biometric authentication [1], emotion identification [2], [3], mental therapy [4], [5], personalised shopping experience [6], [7], security [8], law enforcement and crime prevention, identifying potential security threats in large venues and more. Much research was done during the past 2 to 3 decades on face recognition due to its importance. The initial research works were done in a controlled environment, i.e. a lot of restrictions were imposed for the input face images like controlled lighting, face expression has to be normal, only frontal faces were considered, subjects were not allowed to wear specs during the face image collection, occlusions were not allowed etc. Nowadays, facial recognition is used so widely that we cannot go against all these restrictions. So, identifying faces in an uncontrolled environment becomes unavoidable. The techniques found very effective in a controlled environment fail when the environment becomes uncontrolled. The efficiency of those techniques has declined beyond an acceptable level. With the advancement in artificial intelligence and deep learning (DL), researchers have introduced many techniques to deal with uncontrolled environments [9], [10], [11]. In this paper, one of the challenges faced by face recognition is selected, and techniques that are introduced to overcome that challenge are discussed. The challenge chosen in this paper is occlusion.