Content uploaded by Marc Chaumont
Author content
All content in this area was uploaded by Marc Chaumont on Jan 22, 2018
Content may be subject to copyright.
A preview of the PDF is not available
The emergence of Deep Learning in steganography and steganalysis
Abstract and Figures
A tutorial that was given at a research day in France, the 16th of January 2018. The tutorial is about the infancy of CNNs in steganography/steganalysis in the period 2015 - 2017. Journée ” Stéganalyse : Enjeux et Méthodes”, Poitiers, labelisée par le GDR ISIS et le pré-GDR sécurité.
Figures - uploaded by Marc Chaumont
Author content
All figure content in this area was uploaded by Marc Chaumont
Content may be subject to copyright.
Supplementary resource (1)
... Benefiting from the development of deep learning [37], convolutional neural networks (CNN) perform well in various steganalysis detectors [16], [17], [18], [19], [21], [38], [39], [58]. CNN can automatically extract complex statistical dependencies from images and improve the detection accuracy. ...
... Yedroudj-Net does not use pooling until block 2 and Zhu-Net does not use pooling until layer 4. As report in [54], it is probably better not to use pooling in the first blocks. Because average pooling is a low-pass filter, and it suppresses stego signals by averaging embedding changes in feature maps [58], which is the opposite of our hope that the network can enhance the signal to noise. ...
For steganalysis, many studies showed that convolutional neural network (CNN) has better performances than the two-part structure of traditional machine learning methods. Existing CNN architectures use various tricks to improve the performance of steganalysis, such as fixed convolutional kernels, the absolute value layer, data augmentation and the domain knowledge. However, some designing of the network structure were not extensively studied so far, such as different convolutions(inception, xception, etc.) and variety ways of pooling(spatial pyramid pooling, etc.). In this paper, we focus on designing a new CNN network structure to improve detection accuracy of spatial-domain steganography. First, we use 3×3 kernels instead of the traditional 5×5 kernels and optimize convolution kernels in the preprocessing layer. The smaller convolution kernels are used to reduce the number of parameters and model the features in a small local region. Next, we use separable convolutions to utilize channel correlation of the residuals, compress the image content and increase the signal-to-noise ratio (between the stego signal and the image signal). Then, we use spatial pyramid pooling (SPP) to aggregate the local features and enhance the representation ability of features by multi-level pooling. Finally, data augmentation is adopted to further improve network performance. The experimental results show that the proposed CNN structure is significantly better than other five methods such as SRM, Ye-Net, Xu-Net, Yedroudj-Net and SRNet, when it is used to detect three spatial algorithms such as WOW, S-UNIWARD and HILL with a wide variety of datasets and payloads.
... The search for grey literature included all types of documents contributed by the most relevant researchers in the area. In the presentations of VOLUME 7, 2019 congresses, symposia or short courses exposed in [42], [56]- [60], [80], a good baseline was found to start with a chronological review of the literature. ...
... Regarding the questions set forth in Section II-B we have shown that (1) detection performance obtained by DL applied to steganalysis has surpassed the results obtained by traditional methods (SRM+EC), as can be seen in Figures 7, 8 and Tables 4, 6; (2) a variety of architectures address the specific challenge of steganalysis as shown in Figure 6 and Table 6, furthermore, specific network components have been developed to address this challenge (such as TLU and Gaussian activation functions); (3) current detection levels are those reported in Tables 4, 6, however they are yet far from results targeted by the research community, as conveyed in most of the literature, and specially in [34], [40], [42], [45], [56], [60], [87]; and (4) there is a limited but active set of databases for benchmarking steganalysis methods (see Table 6), however in the new challenge of steganalysis ALASKA [87] new real world oriented databases were released. These databases will serve as a basis for new experiments in steganography and steganalysis. ...
Steganography consists of hiding messages inside some object known as a carrier in order to establish a covert communication channel so that the act of communication itself goes unnoticed by observers who have access to that channel. Steganalysis is dedicated to the detection of hidden messages using steganography; these messages can be implicit in different types of media, such as digital images, video files, audio files or plain text. Traditionally, steganalysis has been divided into two separate stages, the first stage consists of manual extraction of sophisticated features and the second stage is classification using Ensemble Classifiers or Support Vector Machines. In recent years, the development of Deep Learning has made it possible to unify and automate the two traditional stages into an end to end approach with promising results. This paper shows the evolution of steganalysis in recent years using Deep Learning techniques. The results of these techniques have surpassed those obtained with conventional methods -Rich Models with Ensemble Classifiers -both in the spatial and frequency (JPEG) domains. Since 2014, researchers have used Convolutional Neural Networks to solve this problem generating diverse architectures and strategies to improve the detection percentages of steganographic images on the last generation algorithms (WOW, S-UNIWARD, HUGO, J-UNIWARD, among others). Deep Learning, being applied to steganalysis, is now in the process of construction and results so far are encouraging for researchers that are interested in the topic.
... The experimental results show that this framework can improve the safety performance. At present, there is no guarantee [86] that the probability map obtained will defeat the security performance of HILL or S-UNIWARD with STC in practice. It is also unclear whether the loss of the generator must incorporate terms related to safety and terms of payload size. ...
In the past few years, the Generative Adversarial Network (GAN) which proposed in 2014 has achieved great success. GAN has achieved many research results in the field of computer vision and natural language processing. Image steganography is dedicated to hiding secret messages in digital images, and has achieved the purpose of covert communication. Recently, research on image steganography has demonstrated great potential for using GAN and neural networks. In this paper we review different strategies for steganography such as cover modification, cover selection and cover synthesis by GANs, and discuss the characteristics of these methods as well as evaluation metrics and provide some possible future research directions in image steganography.
Today, the most accurate steganalysis
methods for digital media are built as supervised
classifiers on feature vectors extracted from the
media. The tool of choice for the machine learning
seems to be the support vector machine (SVM).
In this paper, we propose an alternative and wellknown
machine learning tool – ensemble classifiers
– and argue that they are ideally suited for
steganalysis. Ensemble classifiers scale much more
favorably w.r.t. the number of training examples
and the feature dimensionality with performance
comparable to the much more complex SVMs. The
significantly lower training complexity opens up the
possibility for the steganalyst to work with rich
(high-dimensional) cover models and train on larger
training sets – two key elements that appear necessary
to reliably detect modern steganographic algorithms.
Ensemble classification is portrayed here
as a powerful developer tool that allows fast construction
of steganography detectors with markedly
improved detection accuracy across a wide range
of embedding methods. The power of the proposed
framework is demonstrated on two steganographic
methods that hide messages in JPEG images.
Nowadays the prevailing detectors of steganographic communication in digital images mainly consist of three steps, i.e., residual computation, feature extraction and binary classification. In this paper, we present an alternative approach to steganalysis of digital images based on convolutional neural network (CNN), which is shown to be able to well replicate and optimize these key steps in a unified framework and learn hierarchical representations directly from raw images. The proposed CNN has a quite different structure from the ones used in conventional computer vision tasks (CVs). Rather than a random strategy, the weights in first layer of the proposed CNN are initialized with the basic high-pass filter set used in calculation of residual maps in Spatial Rich Model (SRM), which acts as a regularizer to suppress the image content effectively. To better capture the structure of embedding signals, which usually have extremely low SNR (stego signal to image content), a new activation function called truncated linear unit (TLU) is adopted in our CNN model. Finally, we further boost the performance of the proposed CNN based steganalyzer by incorporating the knowledge of selection channel. Three state-of-the-art steganographic algorithms in spatial domain, e.g., WOW, S-UNIWARD and HILL are used to evaluate the effectiveness of our model. Compared to SRM and its selection-channel-aware variant maxSRMd2, our model achieves superior performance across all tested algorithms for a wide variety of payloads.
Recent studies have indicated that the architectures of convolutional neural networks (CNNs) tailored for computer vision may not be best suited to image steganalysis. In this letter, we report a CNN architecture that takes into account knowledge of steganalysis. In the detailed architecture, we take absolute values of elements in the feature maps generated from the first convolutional layer to facilitate and improve statistical modeling in the subsequent layers; to prevent overfitting, we constrain the range of data values with the saturation regions of hyperbolic tangent (TanH) at early stages of the networks and reduce the strength of modeling using 1 × 1 convolutions in deeper layers. Although it learns from only one type of noise residual, the proposed CNN is competitive in terms of detection performance compared with the SRM with ensemble classifiers on the BOSSbase for detecting S-UNIWARD and HILL. The results have implied that well-designed CNNs have the potential to provide a better detection performance in the future.
Automatic steganographic distortion learning using a generative adversarial network CNN "simulating" an embedding in a spatial image, CNN called Generator (denoted G) generates the map of modifications
- Tang
[Tang et al. 2017] "Automatic steganographic distortion learning using a generative adversarial
network", W. Tang, S. Tan, B. Li, and J. Huang, IEEE Signal Processing Letter, Oct. 2017
CNN "simulating" an embedding in a spatial image,
CNN called Generator (denoted G) generates the map of
modifications (-1 / 0 / +1),
G learns to embed through a "competition" (GAN methodology)
between it and a Discriminator (noted D).
Generative Adversarial Networks
GAN [Goodfellow 2014] "Generative Adversarial Networks", I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D.
Adaptive Steganography by Oracle (ASO)
- S Kouider
- M Chaumont
- W Puech
ASO [Kouider 2013] "Adaptive Steganography by Oracle (ASO)", S. Kouider and M. Chaumont, W. Puech, ICME'2013.
Technical Points About Adaptive Steganography by Oracle (ASO)
- S Kouider
- M Chaumont
- W Puech
ASO [Kouider 2012] "Technical Points About Adaptive Steganography by Oracle (ASO)", S. Kouider, M. Chaumont, W. Puech,
EUSIPCO'2012.
Deep Learning in "stega
- Chaumont Marc
- Emergence
Marc CHAUMONT
Emergence Deep Learning in "stega"
January 20, 2018
30 / 37