Low Resolution Face Recognition System Based on ESRGAN
Chengkun Song, Zhujiang He, Yinghuai Yu, Zhenni Zhang · 2021 3rd International Conference on Applied Machine Learning (ICAML) · 2021
Low-resolution face recognition is one of the research hotspots of face recognition today. It can be widely used in face recognition in various scenarios, such as identity verification at stations and classroom check-in. The prior art has achieved better performance in ideal scenarios, but detecting low-resolution images will make it difficult to recognize low-resolution human faces, and the accuracy will eventually decrease. This is why improving the accuracy of low-resolution face recognition (LRFR) is still challenging. We have finished the reasearch aim to solve the problem about LRFR. The super-resolution GAN (SRGAN) and enhanced super-resolution GAN (ESRGAN) used in this search. Comparing these methods, we finally obtain a model that can solve the low-precision problem of LRFR. Our system uses super-resolution reconstruction as the preprocessing step of the LRFR problem, and then uses Facenet to recognize the image. These data sets are Wild Face Tag (LFW), YouTube Face Database (YTF), and Wider Face Dataset. The experimental results show that the accuracy of the ESRGAN based on Facenet of the proposed system in the unconstrained natural environment is as high as 98.78%. At the same time, increase the number and speed of face detection, effectively realize the function of multiple face recognition, has practical application value and system robustness.