Mehdi Yedroudj

Hugo Ruiz, Mehdi Yedroudj, Marc Chaumont et al.

For many years, the image databases used in steganalysis have been relatively small, i.e. about ten thousand images. This limits the diversity of images and thus prevents large-scale analysis of steganalysis algorithms. In this paper, we describe a large JPEG database composed of 2 million colour and grey-scale images. This database, named LSSD for Large Scale Steganalysis Database, was obtained thanks to the intensive use of \enquote{controlled} development procedures. LSSD has been made publicly available, and we aspire it could be used by the steganalysis community for large-scale experiments. We introduce the pipeline used for building various image database versions. We detail the general methodology that can be used to redevelop the entire database and increase even more the diversity. We also discuss computational cost and storage cost in order to develop images.

Hugo Ruiz, Marc Chaumont, Mehdi Yedroudj et al.

Since the emergence of deep learning and its adoption in steganalysis fields, most of the reference articles kept using small to medium size CNN, and learn them on relatively small databases. Therefore, benchmarks and comparisons between different deep learning-based steganalysis algorithms, more precisely CNNs, are thus made on small to medium databases. This is performed without knowing: 1. if the ranking, with a criterion such as accuracy, is always the same when the database is larger, 2. if the efficiency of CNNs will collapse or not if the training database is a multiple of magnitude larger, 3. the minimum size required for a database or a CNN, in order to obtain a better result than a random guesser. In this paper, after a solid discussion related to the observed behaviour of CNNs as a function of their sizes and the database size, we confirm that the error's power-law also stands in steganalysis, and this in a border case, i.e. with a medium-size network, on a big, constrained and very diverse database.

Mehdi Yedroudj, Frédéric Comby, Marc Chaumont

Image steganography aims to securely embed secret information into cover images. Until now, adaptive embedding algorithms such as S-UNIWARD or Mi-POD, are among the most secure and most used methods for image steganography. With the arrival of deep learning and more specifically the Generative Adversarial Networks (GAN), new techniques have appeared. Among these techniques, there is the 3 player game approaches, where three networks compete against each other.In this paper, we propose three different architectures based on the 3 player game. The first-architecture is proposed as a rigorous alternative to two recent publications. The second takes into account stego noise power. Finally, our third architecture enriches the second one with a better interaction between the embedding and extracting networks. Our method achieves better results compared to the existing works GSIVAT, HiDDeN, and paves the way for future research on this topic.

Mehdi Yedroudj, Frederic Comby, Marc Chaumont

For about 10 years, detecting the presence of a secret message hidden in an image was performed with an Ensemble Classifier trained with Rich features. In recent years, studies such as Xu et al. have indicated that well-designed convolutional Neural Networks (CNN) can achieve comparable performance to the two-step machine learning approaches. In this paper, we propose a CNN that outperforms the state-ofthe-art in terms of error probability. The proposition is in the continuity of what has been recently proposed and it is a clever fusion of important bricks used in various papers. Among the essential parts of the CNN, one can cite the use of a pre-processing filterbank and a Truncation activation function, five convolutional layers with a Batch Normalization associated with a Scale Layer, as well as the use of a sufficiently sized fully connected section. An augmented database has also been used to improve the training of the CNN. Our CNN was experimentally evaluated against S-UNIWARD and WOW embedding algorithms and its performances were compared with those of three other methods: an Ensemble Classifier plus a Rich Model, and two other CNN steganalyzers.

Mehdi Yedroudj, Marc Chaumont, Frédéric Comby

Deep learning and convolutional neural networks (CNN) have been intensively used in many image processing topics during last years. As far as steganalysis is concerned, the use of CNN allows reaching the state-of-the-art results. The performances of such networks often rely on the size of their learning database. An obvious preliminary assumption could be considering that "the bigger a database is, the better the results are". However, it appears that cautions have to be taken when increasing the database size if one desire to improve the classification accuracy i.e. enhance the steganalysis efficiency. To our knowledge, no study has been performed on the enrichment impact of a learning database on the steganalysis performance. What kind of images can be added to the initial learning set? What are the sensitive criteria: the camera models used for acquiring the images, the treatments applied to the images, the cameras proportions in the database, etc? This article continues the work carried out in a previous paper, and explores the ways to improve the performances of CNN. It aims at studying the effects of "base augmentation" on the performance of steganalysis using a CNN. We present the results of this study using various experimental protocols and various databases to define the good practices in base augmentation for steganalysis.