Illumination Normalization of Face Image Based-on Multi-stage Feature Maps
Shenggui Ling, Keren Fu, Ye Lin, Jiangping Zhu · 2020
Nowadays, illumination normalization of facial image based on Generative Adversarial Networks (GANs) has made noticeable progress. However, the image quality and the recognition accuracy are unsatisfactory when the faces encounter severe illumination. For these reasons, we put forward a new scheme for normalizing face illumination. Moreover, feature maps with multiple sizes are extracted by different convolutional layers of pretrained feature network, and then use these feature maps to compute loss, we call it multi-stage feature maps (MSFM) loss. Qualitative and quantitative experimental results show that the proposed method achieves favorable illumination normalization results against previous models under various illumination challenges.