Research On Defense Method Of Visual Identity Deep Forgery Based On Deep Learning And Compressed Sensing

Nowadays,social networks go deep into everyone’s daily life,and personal information spreads rapidly on the Internet.In particular,the dissemination of visual identity information,such as self photographing and short video,is the most common.However,with the continuous development of deep learning,forged face data has made rapid progress in both authenticity and universality.Deep forgery technology is impacting national security and social stability.Defense and detection of forged visual identity has become a hot research direction.At present,many researchers have studied the defense technology of visual identity deep forgery,and put forward a series of defense methods from different perspectives.According to the different defense stages,the existing defense technologies can be divided into passive detection technology and active defense technology.This paper puts forward new detection methods for these two aspects.Technology is a double-edged sword.Since deep learning technology can be used to generate forged faces,it is also helpful in the detection of forgeries.This paper proposes a passive defense method for synthetic face fusion boundary detection based on deep neural network,which shields the poor detection quality caused by different forgery methods,breaks the limitations of single forgery detection,and improves the generalization of the detection model.Compressed sensing technology is a lossy signal compression method.It can realize data compression and encryption at the same time,which means that using this technology can not only reduce the amount of data transmitted,but also provide a certain security guarantee,and can well meet the needs of visual identity active defense in social networks.The main research results and innovations of this paper are as follows:(1)A human mask model based on full convolution dense network is proposed.The model is a typical encoder decoder structure,which can fully extract and fuse image features in the case of small network scale.On this basis,the generation countermeasure network is used to continuously guide the decoder to generate human mask with better quality.(2)A passive defense method of visual identity based on face mask is proposed.Using this method to judge whether the face in an image has been changed does not need to know the specific operation method of changing the face,but by detecting whether the image is mixed by two images from different sources,that is,detecting the boundary of changing the face.(3)An active defense method of visual identity based on compressed sensing is proposed.Because compressed sensing has the function of encryption,all this technology can prevent the theft and disclosure of visual identity information.However,the encryption degree of compressed sensing technology is weak,and the key scale(the whole measurement matrix)is too large,which leads to its shortcomings in security and space utilization.In view of this,this paper introduces SM3 hash chain algorithm to make the defense method meet the requirements of storage,communication efficiency and security.