The use of convolutional neural networks to enhance the security of steganographic methods

The paper proposes the approach to increase the secrecy of steganography methods in images using convolutional neural networks (CNN). CNNs are integrated into the data embedding process and allow you to minimize the traces of concealment that can be detected by stegoanalyzers. Two implementation options are considered: based on the least significant bit (LSB) and discrete cosine transform (DCT) methods, and also their modifications using CNN. The task of ensuring the secrecy and reliability of embedding was solved in stages: the visual quality of the images, the reliability of message extraction and the resistance to detection by classical methods of stegoanalysis were analyzed. The results of quality and secrecy assessment experiments have confirmed the effectiveness of the proposed approach.