Explicit Optimization of min max Steganographic Game

Abstract

This paper proposes an algorithm which allows Alice to simulate the game played between her and Eve. Under the condition that the set of detectors that Alice assumes Eve to have is sufficiently rich (e.g. CNNs), and that she has an algorithm enabling to avoid detection by a single classifier (e.g adversarial embedding, gibbs sampler, dynamic STCs), the proposed algorithm converges to an efficient steganographic algorithm. This is possible by using a min max strategy which consists at each iteration in selecting the least detectable stego image for the best classifier among the set of Eve's learned classifiers. The algorithm is extensively evaluated and compared to prior arts and results show the potential to increase the practical security of classical steganographic methods. For example the error probability Perr of XU-Net on detecting stego images with payload of 0.4 bpnzAC embedded by J-Uniward and QF 75 starts at 7.1% and is increased by +13.6% to reach 20.7% after eight iterations. For the same embedding rate and for QF 95, undetectability by XU-Net with J-Uniward embedding is 23.4%, and it jumps by +25.8% to reach 49.2% at iteration 3.

Detectability-Based JPEG Steganography Modeling the Processing Pipeline: The Noise-Content Trade-off

“…ment pipeline, which can simply be estimated using only one (Raw,Cover) pair. Furthermore, it requires neither the computationally expensive training and retraining operations of [29], [30] nor the multiple generations of stego databases needed to update the adversary.…”

“…One can after retrain a new classifier and iterate. An efficient strategy to iterate [30] is to select the stego images which are the less detectable using the best trained classifiers. Note that, if the gain is substantial in term of undetectability after more than 5 iterations, this class of embedding schemes is heavy to deploy in practice.…”

“…The proposed scheme is also adversarial in the sense that the embedding is directly designed against an classifier, here the Likelihood Ratio Test. However contrary to adversarial embedding methods based on deep neural nets [29] combined with min max iterations [30], our scheme is only based on the knowledge of the statistical properties of the develop-…”

Image Steganography Based on Iterative Adversarial Perturbations Onto a Synchronized-Directions Sub-Image

“…We directly trained YeNet under payload rate 0.2 bpp instead of adapting curriculum learning. We employed ADV-EMB [14] and MinMax [16] methods for comparison. For MinMax, ADV-EMB were used to produce adversarial stego images.…”

“…For countering CNN steganalystic models [9,10,11] detecting, inspired by adversarial examples [12], there have been some steganographic schemes [13,14,15] proposed, of which ADV-EMB [14] achieves good performance. Furthermore, with min-max strategy, MinMax scheme [16] is proposed to enhance steganographic performance of ADV-EMB.…”