Research Of Image Steganalysis Based On Deep Neural Network

Image steganography is a method of information hiding,which embeds information to normal images secretly.It is showed that spy agencies,terrorist organizations and drug gangs are conducting illegal activities with steganography technology on the Internet,which has been a focus in the field of information security.So researching on image steganalysis technology is urgent and important.JPEG image steganography algorithms in frequency domain such as nsF5 change DCT coefficients of decompressed JPEG images,hiding secret information with little impact on statistical indicators.Classical steganalysis methods for steganography algorithms such as nsF5 mainly depend on the features of manual design,using classifiers such as SVM.With the development of deep learning,Convolutional Neural Networks have been widely used in image classification.In this paper,a novel CNN structure is presented,which can effectively capture tiny artifacts caused by steganography algorithms such as nsF5.Based on the study of steganography,the proposed deep learning network model introduces dense connection blocks,where shortcuts concatenate input features and output features.Retaining features from shallow layers and reducing model parameters,dense connection blocks solve the problem of gradient disappearance in deep networks.Experiments show that the proposed algorithm performs better than traditional methods.