Year: 2025
Type: article
Abstract: The article discusses the possibilities of using deep autoencoders in the field of information hiding (steganography). It is shown that the combination of steganography methods with deep learning makes it possible to improve system reliability and increase the bandwidth of the hidden data transmission channel. The purpose of this paper is to analyse modern approaches to the use of deep autoencoders in steganography, to determine their main advantages and disadvantages, and to formulate promising directions for further research. The development of deep learning opens up broad prospects for enhancing classical steganographic approaches. The use of deep neural networks allows increasing the bandwidth of the steganographic concealment channel, increase resistance to attacks, in particular, to compression, noise and other distortions and adapt to different types of media and specific security requirements will allow identifying new tools for optimising the process of information concealment. The scientific novelty lies in the combination of the ideas of steganography and deep auto-encoders, which makes it possible to obtain high quality recovery of hidden information with minimal visual, acoustic or other artefacts. A comparative review of modern autoencoder architectures is carried out, the principles of encoding and decoding are analysed, and the results of experimental studies demonstrating the effectiveness of the proposed approaches are summarised. The prospects for the development of this area in terms of security, efficiency and resistance to attacks are assessed through a detailed analysis of potential vulnerabilities and practical implementation scenarios. The results of the study indicate the significant potential of deep autoencoders in the field of information security, in particular for integration with steganographic methods. A number of recommendations for further improvement of the technology are proposed, including optimisation of the neural network architecture, expansion of the scope of applications, and consideration of ethical and legal aspects. The results of the study can be used in various areas of information security, including: digital watermarks for copyright protection; covert transmission of messages in communication channels resistant to interception; security of IoT devices, where it is important to minimise the amount of data and at the same time maintain secrecy. The methodology itself can be improved by integrating additional encryption at the level of the hidden message, which will increase the overall level of security.
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4411283099