Research On Image Steganalysis Method Based On Convolutional Neural Network

The traditional steganographic analysis method requires manual design of image features,which requires researchers to spend time designing image features and have research experience in the field of image steganographic analysis.As a result,the traditional steganographic analysis method develops very slowly.To reduce the labor cost of image steganalysis and accelerate the rapid development of image steganalysis,an image steganalysis method based on artificial intelligence(AI)came into being.The AI-based image steganalysis method uses neural network technology to automatically extract features that are conducive to image steganalysis,avoiding the complicated steps of manually extracting features.The AI-based method does not require researchers to have a lot of knowledge in the field of image steganalysis.Researchers of image steganalysis only need to pay attention to the design of neural networks,which accelerates the development of image steganalysis.The AI-based image steganalysis method includes two modules: image noise information extraction and network model design.High-Pass Filter(HPF)and Spatial Rich Model(SRM)are common image noise information extraction methods,but the parameters of these two methods are fixed,which will result in loss of image information.There are researchers who use neural networks to extract noise through automatic learning methods.However,image steganography only modifies a little image pixel and produces less noise information,which is difficult to extract completely through automatic learning methods.The existing image steganalysis model is generally implemented by a single neural network structure,which will cause a high deviation between the detection result and the true value,thereby affecting the performance of the image steganalysis.To improve the performance of image steganalysis,this paper proposes an image noise extraction method and two AI-based image steganalysis models.Aiming at the problem of insufficient image noise information extraction by existing image steganalysis technology,this paper proposes an HPF neural network module to enhance the image noise information.The HPF module not only has the function of the HPF layer,but also can be optimized through model training.To improve the performance of image steganalysis,this paper proposes a novel AI-based image steganalysis model HPF-SRNet.HPF-SRNet uses the HPF module to extract the noise information of the image,and uses a novel nonlinear expression module to increase the model's Non-linear expression ability.Aiming at the problem of high deviations in current AIbased image steganalysis methods,this paper proposes an image steganalysis method EGN based on integrated Google Net.EGN uses variants to learn the deviations of its previous models,and finally makes the entire model The deviation is small.Finally,a large number of experiments were carried out on the commonly used image steganalysis data sets,and the experimental results proved that the HPF module proposed in this paper can effectively extract image noise information.The HPF-SRNet model can not only improve the performance of image steganalysis,but also effectively reduce the training time of the model.The EGN model improves the accuracy of image steganalysis while reducing model deviation.