Research on the Accuracy and Efficiency of Near-infrared Face Recognition
Minrui Yan · 2022 IEEE International Conference on Advances in Electrical Engineering and Computer Applications (AEECA) · 2022
Compared with RGB images, which are easily affected by light, near-infrared imaging face recognition has gradually received widespread attention. Current research mainly focuses on traditionally handcrafted feature extraction algorithms, which have low efficiency and accuracy in near-infrared face recognition. There are also some studies using deep learning methods, but these studies mainly focus on the case of large-size images as input. This paper designed a convolutional neural network (CNN) method for smaller near-infrared images. The experimental results on CBSR NIR Face Dataset show that our proposed CNN method has surpassed other recognition methods such as the Histogram of Oriented Gradient (HOG) and Triplet-Loss in recognition accuracy. Additionally, this paper reduced the size of face images while ensuring high accuracy to improve training efficiency and usage efficiency.