Deep learning is a good steganalysis tool when embedding key is reused for different images, even if there is a cover sourcemismatch

Abstract

Since the BOSS competition, in 2010, most steganalysis approaches use a learning methodology involving two steps: feature extraction, such as the Rich Models (RM), for the image representation, and use of the Ensemble Classifier (EC) for the learning step. In 2015, Qian et al. have shown that the use of a deep learning approach that jointly learns and computes the features, was very promising for the steganalysis.In this paper, we follow-up the study of Qian et al., and show that in the scenario where the steganograph always uses the same embedding key for embedding with the simulator in the different images, due to intrinsic joint minimization and the preservation of spatial information, the results obtained from a Convolutional Neural Network (CNN) or a Fully Connected Neural Network (FNN), if well parameterized, surpass the conventional use of a RM with an EC.First, numerous experiments were conducted in order to find the best "shape" of the CNN. Second, experiments were carried out in the clairvoyant scenario in order to compare the CNN and FNN to an RM with an EC. The results show more than 16% reduction in the classification error with our CNN or FNN. Third, experiments were also performed in a cover-source mismatch setting. The results show that the CNN and FNN are naturally robust to the mismatch problem.In Addition to the experiments, we provide discussions on the internal mechanisms of a CNN, and weave links with some previously stated ideas, in order to understand the results we obtained. We also have a discussion on the scenario "same embedding key".

Yedroudj-Net: An Efficient CNN for Spatial Steganalysis

“…The pre-processed image then feeds the network. Previous studies [2,3] observed that without this preliminary high-pass filter the CNN converges more slowly. This pre-processing largely suppresses the image content, narrows the dynamic range, and thus increases the signal-to-noise ratio between the weak stego signal (if present) and the image signal.…”

“…Inspired by the benefit of diversity [12], and similarly to [17], we use the 30-basic high-pass filters from SRM [12], instead of using only one filter such as [2,3,5], in order to pre-process the input image. Note that the filters kernel values of the preprocessing block, i.e.…”

“…The first attempt to use Deep Learning methods for steganalysis dates back to 2014 [1] with auto-encoders. One year later Qian et al [2] and Pibre et al [3] proposed to use Convolutional Neural Networks. In 2016, the first results, close to those of the state-of-the-art, were obtained with an ensemble of CNNs [4].…”

Large-Scale JPEG Image Steganalysis Using Hybrid Deep-Learning Framework

“…Up to now we used double-compressed images in the experiments. As reported by Pibre et al [23], CNN based steganalyzers can take advantage of seemingly irrelevant subtle patterns to boost their performance. We must eliminate the possibility that our proposed framework makes use of the double compression artifacts to dispel the doubts of the colleagues.…”

“…As a result, even filtered by state-of-the-art steganalytic kernels (e.g. KV kernel used in [21], [23], [24]) the magnitudes of most of the filtered residual elements are still much larger than the corresponding stego noises.…”