Attention Driven Face Recognition, Learning from Human Vision System

This paper proposes a novel face recognition algorithm inspired by Human Visual System (HVS). Firstly, we learn where people look by recording observers’ eye movements when they are viewing face images. We find that the observers are consistent in the regions fixated and such fixated regions are selected as the salient regions. Secondly, we represent the face images by four scales of Local Binary Gabor Patterns (LGBPs) for the salient regions whereas one scale LGBPs for the others, inspired by the fact that fovea of HVS has a higher spatial acuity than the periphery. Thirdly, we integrate the global information of face images in face recognition. The experimental results demonstrate that the proposed method learning from human beings is comparable with those learned with machine learning algorithms, which shows that the characteristics of the HVS provide valuable insights into face recognition.