INCREASING THE EFFICIENCY OF STEGANOGRAPHY THROUGH THE USE OF IMAGE ENHANCEMENT METHODS AND ARTIFICIAL INTELLIGENCE MODELS

DOI: 10.31673/2409-7292.2025.023202

Abstract

The article investigates the problem of increasing the efficiency of steganographic methods through the use of modern
approaches to image enhancement. Particular attention is paid to preprocessing methods, as well as the use of deep neural
networks, such as ESRGAN, U-Net, and SteganoGAN. The results of experiments using adaptive contrast enhancement and
smoothing are presented, which allows increasing the hidden capacity of the container and reducing the probability of detecting
hidden data. The paper investigates the influence of preprocessing methods on the results of steganographic message hiding. It
was experimentally established that preprocessing of images significantly affects the efficiency of LSB steganography. The best
stealth (high PSNR and SSIM) and resistance to JPEG compression was demonstrated by the approach with adaptive texture
segmentation. Conversion to YCbCr also allows increasing stability without losing bandwidth. At the same time, histogram
equalization worsens stability due to increased contrast. Thus, adaptive preprocessing methods are advisable to use to improve
the security and quality of information hiding. A comparison of artificial intelligence models for steganography tasks was
conducted. In the course of the work, artificial intelligence models used in steganography tasks were analyzed. It was found that
the effectiveness of a specific architecture (for example, U-Net or SteganoGAN) significantly depends on the tasks, the type of
input data, the requirements for channel bandwidth, and the available computing resources. It was concluded that the adaptive
use of deep learning and image preprocessing methods allows to increase both the stability of hidden messages and their
invisibility, which is critically important for modern digital steganography.
Keywords: steganography, image enhancement, ESRGAN, deep learning, information protection, neural networks.

References
1. Zhou J., Liu J., et al. "An Image Preprocessing Framework for Steganography". Journal of Visual
Communication and Image Representation, 2020.
2. Wang H., Chen Y. "Color Space Transformations for Enhanced Steganography". IEEE Transactions on
Information Forensics, 2021.
3. Wang X., Yu K., Wu S., et al. "ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks".
ECCV Workshops, 2018.
4. Ronneberger O., Fischer P., Brox T. "U-Net: Convolutional Networks for Biomedical Image Segmentation".
MICCAI, 2015.
5. Zhu J., Kaplan R., Johnson J., Fei-Fei L. "HiDDeN: Hiding Data With Deep Networks". NeurIPS, 2018.
6. Xiao Y., Zhang L., Qian Z. "Robust Steganography via Contrast Limited Adaptive Histogram Equalization".
Signal Processing: Image Communication, 2019.
7. Baluja S. "Hiding Images in Plain Sight: Deep Steganography". NeurIPS, 2017.
8. Liu Y., Huang Y., et al. "A Survey on Deep Learning Based Steganography and Steganalysis". ACM Computing
Surveys, 2021.
9. Tang W., Li Y., et al. "An Overview of Deep-Learning-Based Image Steganography". IEEE Access, 2022.
10. Kim J., Park H. "Lightweight Deep Learning Models for Real-Time Steganography". Pattern Recognition
Letters, 2022.