Research On The Anti-forensics Method Of Face Image Based On Generative Adversarial Network

With the popularity of image processing software,complex computer graphics processing technology can produce realistic computer-generated images.It is difficult to distinguish computer-generated images and natural images with the naked eye alone.The source forensics research on computer-generated image and natural image came into being.However,most of the current forensics methods utilize computer-generated images generated by computer-generated software such as 3D Max,Maya or Photoshop for performance evaluation,and rarely consider the detection capability of computer-generated images generated by the generative adversarial network,Therefore,it is of great significance to realize the research of image source anti-forensics by using the generative adversarial network.This thesis first introduces the research background,significance a nd research status of digital image source forensics technology and anti-forensics technology,and comprehensively analyzes the problems existing in the current work.Secondly,it describes the theories and technologies involved in computer-generated image and natural image source forensics and anti-forensics Research.Finally,aiming at the problem of computer-generated facial image and natural image source identification,two kinds of digital image source anti-forensics algorithms based on the generative adversarial network are proposed,The specific work is as follows:First,a computer-generated facial image anti-forensics method based on the generative adversarial network is proposed.Aiming at the difference of texture,color and style between the computer-generated image and the natural image,a generator and discriminator network architecture,content loss and style loss are designed,which learns a one-way mapping function from computer-generated image to natural image to generating new face image with good visual quality for deceiving the existing computer-generated image detector.Experimental results and analysis demonstrate that the CG facial images regenerated by the proposed anti-forensics scheme can achieve better visual quality compared with those of the existing CG facial image anti-forensics and domain adaptation methods,and it can strike a good balance between visual quality and deception ability.Second,an anti-forensics method based on the two-way conversion of computer-generated facial image and natural facial image based on generative adversarial networks is proposed.Aiming at the difference between the two kinds of images in the generation mechanism,the sensor pattern noise of the image extracted by Gaussian filter is fused into the generative adversarial networks,so that the final generated image not only retains the content of the original domain image but also introduces the noise of another domain image,and realizes the bidirection al mapping between the computer-generated image and the natural image.The experimental results show that the facial image generated by this method has the better visual quality and stronger deception ability comparing the existing anti-forensics method and the bidirectional domain adaptation method.The two anti-forensics schemes proposed in this thesis have defeated the existing image source forensics method to a certain extent,providing new ideas for the research on source forensics of natural image and computer-generated image.