Enhanced Image Steganography Using Neural Style Transfer

This paper presents a novel approach to image steganography by integrating Neural Style Transfer (NST) with multi-layer embedding techniques. Traditional steganography methods often struggle with the fundamental challenges of limited data capacity and image quality degradation when embedding large volumes of data. Also traditional methods fail to accurately retrieve the secret data at the receiving end and the methods are vulnerable to steganalysis tool.Our system transcends these limitations through a three-module design: a multi-layer embedding module based on LSB method over RGB channels, a Neural Style Transfer module for aesthetic processing, and a Style Generative Adversarial Network (StyleGAN) for preservation of image fidelity. The multi-layer embedding scheme boosts data capacity significantly without loss of visual quality, and NST provides an additional layer of security by hiding secret information naturally within stylized transformations. Experimental results demonstrate that our method provides higher embedding capacity compared to traditional methods without loss of imperceptibility through aesthetic processing. The addition of StyleGAN also enhances visual quality of stego images, rendering them steganalysis-immune. Our method is highly promising in applications where secure transmission of large data is necessary, with correct recovery of data without loss of aesthetic quality of the cover image.