The Application Of Machine Learning In Steganography And Steganalysis

Machine Learning is a way to realize Artificial Intelligence,and it allows computers to learn some implicit rules from data automatically.What's more,we may use these rules to predict the data.The application of machine learning in Steganography and Steganalysis helps them to develop faster than ever.For example,we can use Compressive Sensing in Sparse Representation to compress and encode the secret information in Steganography.On the other hand,we can also use it to extract features from the data and use them to classify stego images from the cover images.In Supervised Learning,we can train classifiers like Support Vector Machines,Neural Networks and Ensemble Classifiers using the tagged data.While in Unsupervised Learning,we can use the feature extracted from the images to cluster in between the stego images and cover images.At present,the research on spatial steganography mainly focuses on two aspects: the first is how to improve the embedding capacity without affecting the visual or hearing effect of original carrier;the second is how improve the security of the algorithm so that the secret signal would not be found out easily.While the steganalysis is mainly about judging the unknown carrier to be a secret message carrier or a normal carrier without knowing the steganography algorithms ahead of time.The main contents and the innovations of the paper are summarized as follows:1.For the disadvantages of the traditional spatial steganography algorithm based on LSB that the embedding capacity of cover image is not high and the processes are skipped for secret signals.This paper proposes a reversible data hiding scheme using compressive sensing and random embedding.The algorithm uses compressive sensing techniques to compress and encode the secret images before embedding,and this act will improve the embedding capacity of the cover image.Besides,the steganography algorithm uses the random embedding algorithm to embed the secret bits in the k-LSB bit planes,and this act will minimize the security risk.The experimental results show that the proposed algorithm has the best result in terms of PSNR when embedding with the same amount of secret message.Also,the experimental results show that this algorithm is better than 8D-PVD under the attack of SPAM steganalysis.2.Traditional spatial steganalysis algorithms require a lot of features to train the classifier,and this paper proposed a steganalysis algorithm based on block compressive sensing and a secret signal recovery scheme.The use of compressive sensing in this algorithm is to help improve the accuracy of classification and reduce the dimensions of features needed.Firstly the proposed algorithm uses directional lifting wavelet transform to sparsely represent the images,and then a compressive sensing matrix is generated using a normalized gaussian distribution model.Next,the compressive sensing model is used to sense the directional lifting wavelet coefficients.Lastly,the residual features obtained by sensing matrix are used to reconstruct the original secret signal.The experimental results show that the proposed steganalysis algorithm works and the signal recovery scheme can be used to reconstruct original signal from the stego images.