Research On Image Information Hiding Technology Based On Perceptual Loss And Generative Adversarial Networks

With the continuous development of information technology,information exchange has become more and more convenient,while security has been more and more challenges,related security technology has also received more and more attention.Steganography in information hiding technology can hide secret information in the carrier and transmit without attracting the attention of the third party.Therefore steganography has become a common technique to ensure information security.In this paper,image is used as the carrier signal of steganography,and deep learning model is used for steganography in airspace.The main achievements of this paper include:1.Propose a steganographic scheme of perceptual loss based on feature reconstruction.This scheme refers to the related research results of transfer learning and adopts the common transfer learning model to extract the feature representation of cover image.Then the feature selection operation is carried out to retain the main part of the feature representation,and the secret information is embedded in the reserved feature representation by quantitative operation.Finally,the loss function of the model is used to optimize the stego image.The loss function of the model is divided into two parts: the image quality part and the feature reconstruction part.The experiment results show that the generated stego image has a relatively high image quality,but the steganographic scheme has the disadvantages of slow processing speed and general applicability.2.In order to overcome the shortcomings of the steganography scheme based on perceptual loss,an improved steganography scheme FRGAN(Feature Reconstruction GAN)based on GAN is proposed after introducing Generative Adversarial Networks(GAN)into the scheme.FRGAN as a whole includes a generator,a discriminator and an extractor.The generator generates the corresponding stego image directly from the input cover image and secret information,improves the generalization of steganographic and speeds up the training speed;the discriminator uses mature steganographic analysis scheme to enhance steganographic security;and the extractor uses the transfer learning model of the original steganographic scheme to ensure the effect of embedding and extracting secret information.Compared with the traditional steganographic scheme,FRGAN can better resist the detection of steganographic analysis and has higher steganographic image quality.Compared with steganographic schemes that also use deep learning,FRGAN has higher image quality and better robustness.