Low-Resolution Face Recognition Based On Face Hallucination Method

With the development of big data and artificial intelligence,identity verification methods have gradually changed.Traditional unlocking methods such as keys and fingerprints are gradually being replaced by more intelligent ways such as face recognition,voiceprint recognition,and gait recognition.Among many emerging recognition methods,face recognition methods have achieved the most satisfactory results and have been applied in many scenes of daily life.In today's digital age,applying face recognition to widely deployed monitoring equipment has become necessary for social security.However,faces images captured in many scenes are not ideal for the reasons such as the size is too small or the definition is low due to the shooting distance,weather,or object moving.However,almost of the current face recognition models only focus on high-resolution faces.The recognition effect of these models will not be ideal for the small faces taken by surveillance in public places.The most intuitive way to solve this problem is to reconstruct the low-resolution faces before inputting these images into the face recognition model.Similar to other fields in computer vision,deep learning models have continuously made breakthroughs in face hallucination in recent years.However,current researches generally focus on the visual effects of the reconstruction results.Those methods have not paid attention to the recognition accuracy of the reconstruction results,which makes them have limited application value.In response to the actual needs and current problems which were mentioned above,we try to use the method of face hallucination to solve the actual problem of low-resolution face recognition.Inspired by related work in recent years,we proposed two models and builds a set of models to be applied to the actual identity verification system in this paper.The main contents are as follows:1.We proposed a reference-based super-resolution model for face reconstruction.According to the unique characteristics of the face data,the model applies the reference-based method to the face hallucination task.And we have designed a training procedure for choosing the reference image.Compared with previous methods,our model has advantages in visual effects as well as the recognition accuracy for reconstruction results.2.We proposed a face hallucination model based on identity information.This model combines ideas in several fields such as face recognition,style transfer and superresolution.Identity information has been used during the process of face reconstruction.The model can reconstruct a high resolution face as well as maintain the original identity information.Compared with previous methods,our model has significant advantages in low-resolution face recognition.3.Based on the previous work,we have built an identity verification systems to show the practical value of our model.The face recognition module of the system integrates cutting-edge face recognition methods.At the same time,it can generally maintain a high recognition accuracy rate for face input of different definition,which has huge practical value.Experiments show that the research results of this paper can effectively improve the accuracy of low-resolution face recognition.According to the research route of this paper,there is still much work that can be done.