MI-STEG:A Medical Image Steganalysis Framework Based on Ensemble Deep Learning

Medical image steganography aims to increase data security by concealing patient-personal information as well as diagnostic and therapeutic data in the spatial or frequency domain of radiological images.On the other hand,the discipline of image steganalysis generally provides a classification based on whether an image has hidden data or not.Inspired by previous studies on image steganalysis,this study proposes a deep ensemble learning model for medical image steganalysis to detect malicious hidden data in medical images and develop medical image steganography methods aimed at securing personal information.With this purpose in mind,a dataset containing brain Magnetic Resonance(MR)images of healthy individuals and epileptic patients was built.Spatial Version of the Universal Wavelet Relative Distortion(S-UNIWARD),Highly Undetectable Stego(HUGO),and Minimizing the Power of Optimal Detector(MIPOD)techniques used in spatial image steganalysis were adapted to the problem,and various payloads of confidential data were hidden in medical images.The architectures of medical image steganalysis networks were transferred separately from eleven Dense Convolutional Network(DenseNet),Residual Neural Network(ResNet),and Inception-based models.The steganalysis outputs of these networks were determined by assembling models separately for each spatial embedding method with different payload ratios.The study demonstrated the success of pre-trained ResNet,DenseNet,and Inception models in the cover-stego mismatch scenario for each hiding technique with different payloads.Due to the high detection accuracy achieved,the proposed model has the potential to lead to the development of novel medical image steganography algorithms that existing deep learning-based steganalysis methods cannot detect.The experiments and the evaluations clearly proved this attempt.

Image Steganalysis Based on Deep Content Features Clustering

The training images with obviously different contents to the detected images will make the steganalysis model perform poorly in deep steganalysis.The existing methods try to reduce this effect by discarding some features related to image contents.Inevitably,this should lose much helpful information and cause low detection accuracy.This paper proposes an image steganalysis method based on deep content features clustering to solve this problem.Firstly,the wavelet transform is used to remove the high-frequency noise of the image,and the deep convolutional neural network is used to extract the content features of the low-frequency information of the image.Then,the extracted features are clustered to obtain the corresponding class labels to achieve sample pre-classification.Finally,the steganalysis network is trained separately using samples in each subclass to achieve more reliable steganalysis.We experimented on publicly available combined datasets of Bossbase1.01,Bows2,and ALASKA#2 with a quality factor of 75.The accuracy of our proposed pre-classification scheme can improve the detection accuracy by 4.84%for Joint Photographic Experts Group UNIversal WAvelet Relative Distortion(J-UNIWARD)at the payload of 0.4 bits per non-zero alternating current discrete cosine transform coefficient(bpnzAC).Furthermore,at the payload of 0.2 bpnzAC,the improvement effect is minimal but also reaches 1.39%.Compared with the previous steganalysis based on deep learning,this method considers the differences between the training contents.It selects the proper detector for the image to be detected.Experimental results show that the pre-classification scheme can effectively obtain image subclasses with certain similarities and better ensure the consistency of training and testing images.The above measures reduce the impact of sample content inconsistency on the steganalysis network and improve the accuracy of steganalysis.

A General Linguistic Steganalysis Framework Using Multi-Task Learning

Prevailing linguistic steganalysis approaches focus on learning sensitive features to distinguish a particular category of steganographic texts from non-steganographic texts,by performing binary classification.While it remains an unsolved problem and poses a significant threat to the security of cyberspace when various categories of non-steganographic or steganographic texts coexist.In this paper,we propose a general linguistic steganalysis framework named LS-MTL,which introduces the idea of multi-task learning to deal with the classification of various categories of steganographic and non-steganographic texts.LS-MTL captures sensitive linguistic features from multiple related linguistic steganalysis tasks and can concurrently handle diverse tasks with a constructed model.In the proposed framework,convolutional neural networks(CNNs)are utilized as private base models to extract sensitive features for each steganalysis task.Besides,a shared CNN is built to capture potential interaction information and share linguistic features among all tasks.Finally,LS-MTL incorporates the private and shared sensitive features to identify the detected text as steganographic or non-steganographic.Experimental results demonstrate that the proposed framework LS-MTL outperforms the baseline in the multi-category linguistic steganalysis task,while average Acc,Pre,and Rec are increased by 0.5%,1.4%,and 0.4%,respectively.More ablation experimental results show that LS-MTL with the shared module has robust generalization capability and achieves good detection performance even in the case of spare data.

A Deep Learning Driven Feature Based Steganalysis Approach

The goal of steganalysis is to detect whether the cover carries the secret information which is embedded by steganographic algorithms.The traditional ste-ganalysis detector is trained on the stego images created by a certain type of ste-ganographic algorithm,whose detection performance drops rapidly when it is applied to detect another type of steganographic algorithm.This phenomenon is called as steganographic algorithm mismatch in steganalysis.To resolve this pro-blem,we propose a deep learning driven feature-based approach.An advanced steganalysis neural network is used to extract steganographic features,different pairs of training images embedded with steganographic algorithms can obtain diverse features of each algorithm.Then a multi-classifier implemented as lightgbm is used to predict the matching algorithm.Experimental results on four types of JPEG steganographic algorithms prove that the proposed method can improve the detection accuracy in the scenario of steganographic algorithm mismatch.

Steganalysis Using Fractal Block Codes and AP Clustering in Grayscale Images

This paper presents a universal scheme （also called blind scheme） based on fractal compression and affinity propagation （AP） clustering to distinguish stego-images from cover grayscale images, which is a very challenging problem in steganalysis. Since fractal codes represent the ＂self-similarity＂ features of natural images, we adopt the statistical moment of fractal codes as the image features. We first build an image set to store the statistical features without hidden messages, of natural images with and and then apply the AP clustering technique to group this set. The experimental result shows that the proposed scheme performs better than Fridrich＇s traditional method.

A Novel Universal Steganalysis Algorithm Based on the IQM and the SRM

The state-of-the-art universal steganalysis method,spatial rich model(SRM),and the steganalysis method using image quality metrics(IQM)are both based on image residuals,while they use 34671 and 10 features respectively.This paper proposes a novel steganalysis scheme that combines their advantages in two ways.First,filters used in the IQM are designed according to the models of the SRM owning to their strong abilities for detecting the content adaptive steganographic methods.In addition,a total variant(TV)filter is also used due to its good performance of preserving image edge properties during filtering.Second,due to each type of these filters having own advantages,the multiple filters are used simultaneously and the features extracted from their outputs are combined together.The whole steganalysis procedure is removing steganographic noise using those filters,then measuring the distances between images and their filtered version with the image quality metrics,and last feeding these metrics as features to build a steganalyzer using either an ensemble classifier or a support vector machine.The scheme can work in two modes,the single filter mode using 9 features,and the multi-filter mode using 639 features.We compared the performance of the proposed method,the SRM and the maxSRMd2.The maxSRMd2 is the improved version of the SRM.The simulated results show that the proposed method that worked in the multi-filter mode was about 10%more accurate than the SRM and maxSRMd2 when the data were globally normalized,and had similar performance with the SRM and maxSRMd2 when the data were locally normalized.

Binary Image Steganalysis Based on Distortion Level Co-Occurrence Matrix

In recent years,binary image steganography has developed so rapidly that the research of binary image steganalysis becomes more important for information security.In most state-of-the-art binary image steganographic schemes,they always find out the flippable pixels to minimize the embedding distortions.For this reason,the stego images generated by the previous schemes maintain visual quality and it is hard for steganalyzer to capture the embedding trace in spacial domain.However,the distortion maps can be calculated for cover and stego images and the difference between them is significant.In this paper,a novel binary image steganalytic scheme is proposed,which is based on distortion level co-occurrence matrix.The proposed scheme first generates the corresponding distortion maps for cover and stego images.Then the co-occurrence matrix is constructed on the distortion level maps to represent the features of cover and stego images.Finally,support vector machine,based on the gaussian kernel,is used to classify the features.Compared with the prior steganalytic methods,experimental results demonstrate that the proposed scheme can effectively detect stego images.

Passive Steganalysis Based on Higher Order Image Statistics of Curvelet Transform

Steganographic techniques accomplish covert communication by embedding secret messages into innocuous digital images in ways that are imperceptible to the human eye. This paper presents a novel passive steganalysis strategy in which the task is approached as a pattern classification problem. A critical part of the steganalyser design depends on the selection of informative features. This paper is aimed at proposing a novel attack with improved performance indices with the following implications： 1） employing higher order statistics from a curvelet sub-band image representation that offers better discrimination ability for detecting stego anomalies in images, as compared to other conventional wavelet transforms; 2） increasing the sensitivity and specificity of the system by the feature reduction phase; 3） realizing the system using an efficient classification engine, a neuro-C4.5 classifier, which provides better classification rate. An extensive experimental evaluation on a database containing 5600 clean and stego images shows that the proposed scheme is a state-of-the-art steganalyser that outperforms other previous steganalytic methods.

A HEVC Video Steganalysis Algorithm Based on PU Partition Modes

Steganalysis is a technique used for detecting the existence of secret information embedded into cover media such as images and videos.Currently,with the higher speed of the Internet,videos have become a kind of main methods for transferring information.The latest video coding standard High Efficiency Video Coding(HEVC)shows better coding performance compared with the H.264/AVC standard published in the previous time.Therefore,since the HEVC was published,HEVC videos have been widely used as carriers of hidden information.In this paper,a steganalysis algorithm is proposed to detect the latest HEVC video steganography method which is based on the modification of Prediction Units(PU)partition modes.To detect the embedded data,All the PU partition modes are extracted from P pictures,and the probability of each PU partition mode in cover videos and stego videos is adopted as the classification feature.Furthermore,feature optimization is applied,that the 25-dimensional steganalysis feature has been reduced to the 3-dimensional feature.Then the Support Vector Machine(SVM)is used to identify stego videos.It is demonstrated in experimental results that the proposed steganalysis algorithm can effectively detect the stego videos,and much higher classification accuracy has been achieved compared with state-of-the-art work.

A BLIND AUDIO STEGANALYSIS BASED ON FEATURE FUSION

In this paper, we present a blind steganalysis based on feature fusion. Features based on Short Time Fourier Transform (STFT), which consists of second-order derivative spectrum features of audio and Mel-frequency cepstrum coefficients, audio quality metrics and features on linear prediction residue are extracted separately. Then feature fusion is conducted. The performance of the proposed steganalysis is evaluated against 4 steganographic schemes: Direct Sequence Spread Spectrum (DSSS), Quantization Index Modulation (QIM), ECHO embedding (ECHO), and Least Significant Bit em-bedding (LSB). Experiment results show that the classifying performance of the proposed detector is much superior to the previous work. Even more exciting is that the proposed methodology could detect the four steganography, with 85%+ classification accuracy achieved in all the detections, which makes the proposed steganalysis methodology capable of being regarded as a blind steganalysis, and especially useful when the steganalyzer are without the knowledge of the steganographic scheme employed in data embedding.

A Survey on Different Feature Extraction and Classification Techniques Used in Image Steganalysis

Steganography is the process of hiding data into public digital medium for secret communication. The image in which the secret data is hidden is termed as stego image. The detection of hidden embedded data in the image is the foundation for blind image steganalysis. The appropriate selection of cover file type and composition contribute to the successful embedding. A large number of steganalysis techniques are available for the detection of steganography in the image. The performance of the steganalysis technique depends on the ability to extract the discriminative features for the identification of statistical changes in the image due to the embedded data. The issue encountered in the blind image steganography is the non-availability of knowledge about the applied steganography techniques in the images. This paper surveys various steganalysis methods, different filtering based preprocessing methods, feature extraction methods, and machine learning based classification methods, for the proper identification of steganography in the image.

Steganalysis of Low Embedding Rate CNV-QIM in Speech

To address the difficulty of detecting low embedding rate and high-concealment CNV-QIM(complementary neighbor vertices-quantization index modulation)steganography in low bit-rate speech codec,the code-word correlation model based on a BiLSTM(bi-directional long short-term memory)neural network is built to obtain the correlation features of the LPC codewords in speech codec in this paper.Then,softmax is used to classify and effectively detect low embedding rate CNV-QIM steganography in VoIP streams.The experimental results show that for speech steganography of short samples with low embedding rate,the BiLSTM method in this paper has a superior detection accuracy than state-of-the-art methods of the RNN-SM(recurrent neural network-steganalysis model)and SS-QCCN(simplest strong quantization codeword correlation network).At an embedding rate of 20%and a duration of 3 s,the detection accuracy of BiLSTM method reaches 75.7%,which is higher than that of RNNSM by 11.7%.Furthermore,the average testing time of samples(100%embedding)is 0.3 s,which shows that the method can realize real-time steganography detection of VoIP streams.

Color Image Steganalysis Based on Residuals of Channel Differences

This study proposes a color image steganalysis algorithm that extracts highdimensional rich model features from the residuals of channel differences.First,the advantages of features extracted from channel differences are analyzed,and it shown that features extracted in this manner should be able to detect color stego images more effectively.A steganalysis feature extraction method based on channel differences is then proposed,and used to improve two types of typical color image steganalysis features.The improved features are combined with existing color image steganalysis features,and the ensemble classifiers are trained to detect color stego images.The experimental results indicate that,for WOW and S-UNIWARD steganography,the improved features clearly decreased the average test errors of the existing features,and the average test errors of the proposed algorithm is smaller than those of the existing color image steganalysis algorithms.Specifically,when the payload is smaller than 0.2 bpc,the average test error decreases achieve 4%and 3%.

An Adaptive Image Calibration Algorithm for Steganalysis

In this paper,a new adaptive calibration algorithm for image steganalysis is proposed.Steganography disturbs the dependence between neighboring pixels and decreases the neighborhood node degree.Firstly,we analyzed the effect of steganography on the neighborhood node degree of cover images.Then,the calibratable pixels are marked by the analysis of neighborhood node degree.Finally,the strong correlation calibration image is constructed by revising the calibratable pixels.Experimental results reveal that compared with secondary steganography the image calibration method significantly increased the detection accuracy for LSB matching steganography on low embedding ratio.The proposed method also has a better performance against spatial steganography.

General Steganalysis Method of Compressed Speech Under Different Standards

Analysis-by-synthesis linear predictive coding(AbS-LPC)is widely used in a variety of low-bit-rate speech codecs.Most of the current steganalysis methods for AbS-LPC low-bit-rate compressed speech steganography are specifically designed for a specific coding standard or category of steganography methods,and thus lack generalization capability.In this paper,a general steganalysis method for detecting steganographies in low-bit-rate compressed speech under different standards is proposed.First,the code-element matrices corresponding to different coding standards are concatenated to obtain a synthetic code-element matrix,which will be mapped into an intermediate feature representation by utilizing the pre-trained dictionaries.Then,bidirectional long short-term memory is employed to capture long-term contextual correlations.Finally,a code-element affinity attention mechanism is used to capture the global inter-frame context,and a full connection structure is used to generate the prediction result.Experimental results show that the proposed method is effective and better than the comparison methods for detecting steganographies in cross-standard low-bit-rate compressed speech.

An Efficient Steganalysis Model Based on Multi-Scale LTP and Derivative Filters

Local binary pattern(LBP)is one of the most advanced image classification recognition operators and is commonly used in texture detection area.Research indicates that LBP also has a good application prospect in steganalysis.However,the existing LBP-based steganalysis algorithms are only capable to detect the least significant bit(LSB)and the least significant bit matching(LSBM)algorithms.To solve this problem,this paper proposes a steganalysis model called msdeLTP,which is based on multi-scale local ternary patterns(LTP)and derivative filters.The main characteristics of the msdeLTP are as follows:First,to reduce the interference of image content on features,the msdeLTP uses derivative filters to acquire residual images on which subsequent operations are based.Second,instead of LBP features,LTP features are extracted considering that the LTP feature can exhibit multiple variations in the relationship of adjacent pixels.Third,LTP features with multiple scales and modes are combined to show the relationship of neighbor pixels within different radius and along different directions.Analysis and simulation show that the msdeLTP uses only 2592-dimensional features and has similar detection accuracy as the spatial rich model(SRM)at the same time,showing the high steganalysis efficiency of the method.

An Effective Steganalysis Algorithm for Histogram-Shifting Based Reversible Data Hiding

To measure the security for hot searched reversible data hiding(RDH)technique,especially for the common-used histogram-shifting based RDH(denoted as HS-RDH),several steganalysis schemes are designed to detect whether some secret data has been hidden in a normal-looking image.However,conventional steganalysis schemes focused on the previous RDH algorithms,i.e.,some early spatial/pixel domain-based histogram-shifting(HS)schemes,which might cause great changes in statistical characteristics and thus be easy to be detected.For recent improved methods,such as some adaptive prediction error(PE)based embedding schemes,those conventional schemes might be invalid,since those adaptive embedding mechanism would effectively reduce the embedding trace and thus increase the difficulty of steganalysis.Therefore,a novel steganalysis method is proposed in this paper to detect recent adaptive RDH schemes and provide a more effective detection tool for RDH.The contributions of this paper could be summarized as follows.(1)By analyzing the characteristics for those adaptive HS-RDH,an effective“flat ground”based detection method is designed to fast identify whether the given image is used to hide secret data;(2)According to the empirical statistical model,double check mechanism is provided to improve the detection accuracy;(3)In addition,to further improve detection ability,some detailed information for secret data,i.e.,its content and embedding location are further estimated.Compared with conventional steganalysis methods,experimental results indicate that our proposed algorithm could achieve a better detection accuracy and meanwhile acquire more detailed information on secret data.

Image Steganalysis System optimization Based on Boundary Samples

In the image steganalysis,the training samples often determine the performance of the model when the features and classification are in the same condition.However the existing research on steganalysis lacks the in-depth study of the classifier's training method which may deeply influence the detection performance.This paper provides an optimization of universal steganalysis based on the boundary samples classification concerning about image steganalysis.This paper proposes a strategy of selecting boundary samples in steganalysis and divides the training samples into good samples,poor samples and boundary samples three categories and then chose the optimal threshold to get boundary samples through experiments.The experimental results show the effectiveness of boundary sample,which dramatically improve detection capability especially for the low embedding rate Stego-image.

A Two-Stage Highly Robust Text Steganalysis Model

With the development of natural language processing,deep learning,and other technologies,text steganography is rapidly developing.However,adversarial attack methods have emerged that gives text steganography the ability to actively spoof steganalysis.If terrorists use the text steganography method to spread terrorist messages,it will greatly disturb social stability.Steganalysis methods,especially those for resisting adversarial attacks,need to be further improved.In this paper,we propose a two-stage highly robust model for text steganalysis.The proposed method analyzes and extracts anomalous features at both intra-sentential and inter-sentential levels.In the first phase,every sentence is first transformed into word vectors.To obtain a high dimensional sentence vector,we use Bi-LSTM to obtain feature information for all words in the sentence while retaining strong correlations.In the second phase,we input multiple sentences vectors into the GNN,from which we extract inter-sentential anomaly features and make a judgment as to whether the text contains secret messages.In addition,to improve the robustness of the model,we add adversarial examples to the training set to improve the robustness and generalization of the steganalysis model.Theoretically,our proposed method is more robust and more accurate in detection compared to existing methods.

A Hierarchical Learning Framework for Steganalysis of JPEG Images

JPEG Steganalysis is an important technique for forensic analysis of images on online social networks. This paper proposes a novel hierarchical learning framework for JPEG steganalysis. It is based on the observation that both regions of an image with different textural complexity and regions of different images with similar textural complexity tend to have different embedding probabilities. In the training stage of our framework, images are firstly clustered into a number of categories using Gaussian Mixture Model (GMM). Then, images in each category are decomposed into smaller blocks, and these blocks are also clustered into limited classes. Finally, a classifier is trained for each class of blocks.In the testing stage, an image and its blocks are also classified using trained GMM, and each block is tested on corresponding classifiers to make the final decision by weighed sum of individual results. Extensive experimental results show a better performance of our framework compared with some other previous learning framework.