A Survey on Occluded Face recognition

Abstract

The human face is a research hotspot in the existing biometric technology. The collection of the faces is affected by unpredictable factors in reality, such as expression change, posture change, illumination, occlusion, and so on, which will reduce the performance of the face recognition system. Occluded face recognition is a long-standing and challenging problem. There are scanty studies on occlusion issues compared with other problems. Occlusion face recognition has profound application potential at present, so it is imperative to deal with the occlusion problem effectively. The first primary problem of face recognition is faced with face detection, so we introduce general face detection and occluded face detection. For general face detection, we briefly describe the methods based on template matching and deep learning. Then we analyze the existing methods of occluded face detection from two aspects, which are 1) based on manual design features and 2) based on deep learning. Secondly, we focus on occluded face recognition from the perspectives of 2D and 3D. With regard to 2D face images, we make an analysis from two categories in detail, which are 1) robust feature extraction and 2) subspace regression. 3D scanners and other 3D acquisition technologies are developing rapidly, and 3D face images contain rich information inherent in the face itself and are robust to occlusion. 3D face recognition is inevitable to become the mainstream in the field of face recognition. Therefore, we analyze and summarize the three popular aspects, which are 1) based on surface information, 2) based on local surface information, and 3) based on deep learning. Finally, we analyze the current problems and future research directions.