Design And Implementation Of The Robust Face Recognition System Against Face Editing

With the development of deep learning techniques,many face editing algorithms based on Generative Adversarial Networks(GANs)have emerged and are widely used in a variety of face editing software.These software can edit the input face image,including adding or removing certain facial features,changing age,gender,etc.Face editing technology may be used for criminal purposes,for example,criminals may upload edited face images or disguise themselves to evade capture.While existing face recognition algorithms perform well on unedited face images,the accuracy rate is not high in the edited face images.At the same time,existing face recognition systems often consider using face editing to augment the face dataset due to the difficulty of collecting face data,but synthetic faces can cause damage to the performance of the model.To address these two issues,this paper proposes a face recognition algorithm based on contrast learning and feature decoupling,which is able to filter out synthetic modalities in the training phase and also accurately recognize edited face images in the testing phase.This algorithm conforms to the properties of good representation learning and is able to learn robust face representations,firstly by liberating face representations from the fixed region of low-dimensional manifolds through contrast learning,and then using feature decoupling to filter out some representations in face representations that are irrelevant to identity information and more tightly clustered in low-dimensional manifolds,resulting in high cohesion of face picture representations of the same identity and low coupling.Based on this algorithm,this paper designs and implements a robust face recognition system against face editing.This paper describes in detail the requirements analysis and the outline design of each module of the system,and then elaborates on the principles and implementation of the data generation module,the face model construction and training module,face model recognition module the effectiveness analysis module,and the visualization module.