Research On Carrier-enhanced Information Hiding Metho

Currently,the development of information intelligence has greatly promoted the growth of information in the network,and the security of information transmission has been widely emphasized.Information hiding technology uses multimedia covers to hide secret messages,aiming to achieve high-security transmission without arousing suspicion from malicious monitoring parties.Existing information hiding methods typically achieve distortionminimizing embedding by designing distortion cost functions to ensure security.However,these methods all directly use the given cover image as the embedding object,ignoring the fact that the embedding ability of the cover image itself will directly affect the security and embedding capacity of information hiding.Therefore,purely pursuing optimization of the distortion cost function still limits by the trade-off between security and embedding capacity,and it is difficult to break through the bottleneck of simultaneous improvement of security and capacity.In addition,although cover selection and batch steganography methods can alleviate the above problems to some extent,selecting cover images with smaller distortions after hiding secret messages is still limited by the inherent distribution of the cover image itself.Considering the contextual semantic environment of covert communication,it is still difficult to achieve optimal embedding performance when a specific cover image is given.To address the above problems,this paper proposes cover-enhancement information hiding methods.Firstly,focusing on the security risk of the whole cover image as the embedding object,the embedding cost is calculated to achieve security enhancement by actively locating the region on the cover image suitable for embedding messages as the embedding object.Considering that the cover image is limited by its own hiding capacity,the content of the cover is modified from the perspective of actively enhancing the hiding capacity,so that the enhanced cover has high hiding capacity and low detectability,thus achieving simultaneous enhancement of hiding capacity and security.In addition,for cover images with specific meaning but not suitable for enhancement,the decoupled embedded dual-tags is proposed to actively verify the authenticity and traceability of cover images respectively,so as to achieve reliable cover image protection.The main research content and innovation of this paper are as follows:Firstly,targeting the issue of the uncertainty of embedding probability may lead to unreasonable embedding,steganography scheme based on cover embedding region localization is proposed.Since the distortion minimization framework is based on probabilistic design,the uncertain embedding progress cannot guarantee that secret messages is embedded in low embedding cost regions,which may lead to the embedding of secret messages in inappropriate regions.To solve this problem and combined with the fact that the background regions of practical cover images are extremely smooth and unsuitable for hiding secret messages,this paper proposes steganography methods based on cover embedding region localization.For algorithmic efficiency,object detection methods are adopted to actively locate rectangle regions for embedding and combine multiple distortion functions to achieve secure embedding.For high-security,considering that there are still smooth background problems in the located region,this paper further proposes to use the instance segmentation method to obtain irregular embedding target region and design an embedding fully convolutional network to embed secret messages precisely in the located region,effectively improving the security of the algorithm.The experiments show that the proposed method can effectively improve the anti-steganalysis performance of stego images,thus improving the security of the hiding methods.Secondly,targeting the issue of insufficient hiding ability in the foreground region of cover images,a steganography method based on foreground enhancement is proposed.The previous research of steganography scheme based on cover embedding region localization embeds secret messages into the localized region,but since the localized region cannot guarantee that all meet secure steganography.Thus,from the perspective of actively improving the hidden ability of the foreground region,this paper proposes a cover texture enhancement module using a style transfer method based on a generative adversarial network.In the process of object texture enhancement,using the network of cyclic cross-domain generator and designing a cycleconsistency loss to ensure the quality of the enhanced cover image.In addition,while realizing texture enhancement,the secret messages embedding process is simulated,and a steganalysis discriminator is added to provide gradients.It ensures enhanced cover images and stego images can anti-steganalysis.Extensive experimental results show that the security of stego images is substantially improved with the same capacity thanks to the effective enhancement of the embedding capability of the target region of the cover image.Thirdly,targeting the issue of insufficient hiding ability in the background region of cover images,a steganography method based on background region enhancement is proposed.In order to further enhance the hiding ability of cover images,to meet higher embedding capacity while ensuring the anti-steganalysis performance,and to address the problem of insufficient hiding capacity in the background region of cover images,from the perspective of actively increasing the secure region,firstly,the hiding ability of each pixel in the cover image is evaluated,and then the high embedding cost region is located,and a foreground object with complex texture and rich noise is generated in that region.In addition,the image quality discriminator is used to ensure the image quality of the enhanced cover image.Compared with the cover selection methods,the hiding ability of the unsuitable hiding region is increased by adjusting the distribution of the cover,that is,generating a suitable object region for embedding secret messages in the high embedding cost region,which enhances the embedding capacity while ensuring the same-level anti-detection performance of the stego images.Extensive experiments demonstrated that the proposed method can exponentially increase the embedding capacity of the cover image and improve the anti-steganalysis performance of the stego image.The proposed cover background object addition scheme can be applied to arbitrary steganography algorithms and cover images at different capacities and certain generalization.Fourthly,targeting the issue of verifying the authenticity of the cover image content,a cover image protection method based on a decoupled invertible network by embedding dual-tags is proposed.This paper focuses on the hiding ability of the cover image itself,by enhancing the cover image,which can effectively achieve simultaneous improvement of security and hiding capacity.However,if the cover image has specific meanings and is not suitable to be enhanced,how to achieve effective protection of the cover image content and verify its content authenticity is a major problem of cover enhancement steganography.Motived by this,this paper proposes to embed traceable and authenticable tags to cover images by designing a decoupled invertible neural network.As for tagged images,different types of attack methods are designed,including moderate and malicious attacks.Besides,it is ensured that the extraction error rates of the dual-tags show robust or fragile performance against different types of attacks by designing the mapping-aware network and distance metric module,so as to achieve both cover content authentication and source verification for reliable cover image protection.