Research Of Detection Methods For JPEG Image Robust Steganography

Steganography is the process of embedding a secret message in a cover and sending it to the receiver to achieve secret communication.Social platform is an important channel for digital steganography,but social network will process digital images lossily,which leads to the failure of traditional steganography methods,therefore,robust steganography that can resist platform processing is the current hot spot of steganography.However,there is a lack of steganalysis methods for targeted robust steganography.In this paper,we propose two targeted steganalysis methods from different perspectives to improve the accuracy of detecting robust steganography.Current robust steganography algorithms can be divided into two categories according to the method of achieving robustness:carrier preprocessing and carrier selection,which have different impacts on security.In this paper,we propose a new detection method for the characteristics of these two types of methods,which can effectively improve the detection accuracy.The main work and innovation points are as follows:1.Propose a detection method based on block featuresThe robust steganography of the cover-preprocessing class mainly use repetitive compression as preprocessing to make the image and channel match,when the modification causes a small compression loss.In this paper,from the theory of JPEG compression process,we can find that the error brought before and after each compression has different changes as the number of compressions increases.The mean and variance are calculated for different errors(i.e.truncation errors and rounding errors)under different transform domains and different coefficients to obtain the error block characteristics.Then the generic detection method proposed in this paper combines the error block features with the traditional steganalysis features and then classifies them by a classifier.For different steganographic features,the method in this paper can be used,and we also perform feature dimensionality expansion due to the randomness of the ensemble classifier.Experiments show that the method in this paper has good generalizability,and the fused features after using different features combined can be better and improve the steganalysis detection accuracy.2.Propose a detection method based on selected channelsThe robust steganography of selection class mainly screens the regions in the image with JPEG compression resistance as the embedding domain after preprocessing for embedding.Although this type of method is relatively easy to be detected by steganalysis,the detection error rate is still high in the case of low embedding rate,which is still unacceptable in the actual detection task with large data volume.However,being able to further improve at low embedding rates can have a greater impact in actual big data detection tasks.As a steganographer,the embedding domain needs to consider whether it is robust and also whether it is suitable for modification.In order to solve the problem of low correct rate of traditional steganalysis detection of this class of robust steganography under low embedding rate.In this dissertation,we propose a detection method based on the selection channel,which considers both distortion and robustness of the modification point,and combines the robust estimation probability and the distortion modification probability as the selection channel.For this purpose,coverage experiments,performance analysis experiments and false alarm rate experiments are done in this dissertation.The experiments show that the method can improve the correct detection rate of this class of robust steganography at low embedding rates,especially the false alarm rate.And it has some generalization property.