Research On Image Steganalysis Algorithms Based On Deep Learning

Nowadays,with the rapid development of network information technology,more attention has been paid to the security of information in the communication process.Steganography,as an effective means of information security transmission,embeds the secret information that needs to be transmitted into the carrier,and then transmits it to the receiving end through the public channel without loss,thus ensuring the security of secret information in the transmission process.However,this technology will also be used by people with ulterior motives,threatening the security and stability of society.In this context,steganalysis technology plays a very important role.As the name implies,steganalysis technology,a technology to analyze whether the carrier in the public channel contains secret information,is opposite to steganalysis technology.Steganographic analysis technology mainly solves the problem of two classifications,that is,to distinguish whether the carrier contains secret information or not.At present,steganalysis technology is mainly divided into two categories:traditional steganalysis technology and deep learning-based steganalysis technology.Traditional steganalysis technology is mainly based on the characteristics of manual design,and then classifies them.However,this method relies too much on the experience of designers and has some limitations.The object of this paper is digital image.The feature learning and classification of Steganalysis based on deep learning are studied.The contents of this paper include:(1)Steganographic analysis of directed multi-filter kernel deep learning network.This work analyses the filter core in the steganalysis method of the rich model,and combines the knowledge of steganography.In the pretreatment layer of the deep learning network,the directed multi-filter core is used.This not only retains the function of the pretreatment layer to increase the signal-to-noise ratio,but also points out the direction for the learning of network parameters.A deep learning network to solve the steganalysis problem is proposed.The structure improves the detection performance of the network.The experimental results show that the directional filtering kernel preprocessed deep learning network can learn the image features directionally.In the case of high embedding rate,the detection effect is better than the spatial rich model,and the detection accuracy is improved compared with some existing deep learning steganalysis methods.(2)Research on Enhanced Differential Transfer Steganalysis in-depth Learning Framework.This work mainly analyses DenseNet and ResNet network's dense connection mode,and considers that the convolution operation in the network may filter out the features that are beneficial to classification,connect the residuals of the former layer or the former layer after convolution to the latter layer,so as to prevent the loss of useful information.A new framework for in-depth learning of enhanced differential transmission steganalysis is is proposed.Step by step to improve the accuracy of network classification.By adding the operation of enhancing differential transmission to the network,the validity of this work to the accuracy of classification is proved.