FEATURES OF APPLICATION OF PRETRAINED CONVOLUTIONAL NEURAL NETWORKS TO GRAPHIC IMAGE STEGOANALYSIS
S. N. Tereshchenko1, A. A. Perov2, A. L. Osipov1
1Novosibirsk State University of Economics and Management, Novosibirsk, Russia
2Moscow Polytechnic University, Moscow, Russia
Keywords: machine learning, convolutional neural networks, steganography, steganalysis, container
Abstract
The use of convolutional neural networks for analyzing graphic images for the presence of data introduced by steganography methods is investigated. It is shown that a deep convolutional neural network is trained to classify the presence of hidden data in graphic images, achieving accuracy according to the weighted AUC metric of 0.928. The hypothesis of the effectiveness of applying the concept of "transfer learning" for the sphere of steganography is tested. The effectiveness of the proposed technology is confirmed by a large number of experiments.
|
