A Method of Dual-spectrum Feature Fusion for Face Recognition Under Non-ideal Lighting Conditions
Da Ai, Weixin Fan, Kai Jia, Mingyue Lu, Ying Liu · 2022
Face recognition technology is widely used in the field of public security. To improve the recognition accuracy under non-ideal lighting conditions, a face recognition method with dual-spectrum feature fusion is proposed using the property that the infrared spectrum is insensitive to visible light. The fused face images of visible and near-infrared spectra are obtained with the Non-Subsampled Shearlet Transform (NSST) algorithm, and then been put into the FaceNet as input and trained using transfer learning to renew the FaceNet model parameters for recognizing the fused face images. Compared with existing methods, experimental results show that the accuracy of face recognition is significantly improved under non-ideal lighting conditions which better meets the practical application requirements of public security.