Research And Application Of Image Steganography Forensics Based On Adversarial Learning

With the rapid development of network and digital technology,the urgent needs of covert communication and digital copyright protection make the ancient steganography glow with new vitality in the network era.Image steganography is a technology that hides secret information into images.Due to its simplicity,difficulty of detection and extraction,it can be easily used by illegal organizations to engage in activities that will endanger national security.This thesis mainly studies the detection of steganography,also known as image steganography forensics.In recent years,the academic community has proposed a variety of steganalysis methods based on deep learning,using convolutional neural networks to automatically extract classification features and achieve good detection performance.However,convolutional neural networks can effectively extract the features describing the image content,the difficulty lies in extracting the subtle features that describe the existence of hidden information.Moreover,in order to solve the difficulty of convergence in steganalysis using traditional networks,a high pass filter is introduced to process the input,but the design of the high pass filter is not necessarily optimal,which may suppress some stego signals.In order to solve the problems in the previous methods,this thesis proposes an image steganography forensics method based on adversarial learning.Using the deep neural network as the tool and introducing the adversarial learning mechanism,a deep neural network structure of steganography detection based on adversarial learning is established.Based on the existing methods,the proposed method further extracts the features that describe the existence of hidden information and improves the signal-to-noise ratio,so as to realize the accurate classification of cover image and steganographic image.The specific research contents are as follows:(1)This thesis focuses on the problem of image steganography forensics,introduces separable convolution and adversarial learning,and proposes a new algorithm for image steganalysis,which not only improves the efficiency,but also achieves good detection performance.(2)A color image steganography forensics algorithm based on the above framework is proposed.By introducing separate channel-wise convolution and diverse activation modules to enhance the signal-to-noise ratio,the algorithm extracts more features that describe the existence of hidden information that are beneficial to steganalysis,and improves the accuracy of color image steganography detection.(3)The algorithm proposed in this thesis is applied to the actual scene from theory,and the corresponding image steganography detection system is designed and developed.The visual interface is provided for users to realize the steganography detection and analysis of real images.