Improved image steganography system using convolutional neural networks.
Date
2024-11
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Federal University of Technology, Owerri
Abstract
This study presents an innovative image steganography method that combines convolutional autoencoders with the Residual Network (ResNet) architecture to address the limitations of traditional techniques. The method aims to securely embed data within images while maintaining high imperceptibility and efficiency. Existing methods struggle with balancing image quality, embedding capacity, and computational complexity, highlighting the need for advanced solutions. This approach enables the concealment of color images within others, ensuring secure data transmission and reducing computational complexities. The method utilizes ResNet-based preprocessing for feature extraction and embedding operations, evaluated using the CIFAR dataset (60,000 images). Experiments were conducted using Python programming, TensorFlow, and NVIDIA GPUs for training and testing. The model achieves PSNR values exceeding 30 dB and SSIM values above 0.98, ensuring superior imperceptibility, reduced complexity, and effective image quality retention. This deep-learning-based approach offers a significant improvement in security, embedding capacity, imperceptibility, and computational efficiency, advancing the field of image steganography.
Description
Master thesis on ".Improved image steganography system"
Keywords
Image steganography, deep neural networks, autoencoder, ResNet, CIFAR dataset, computational efficiency., computer science
Citation
Ofoegbu, C. I. (2024). Improved image steganography system using convolutional neural networks. (Unpublished Master's Thesis), Federal University of Technology, Owerri