Image Steganography: A Comprehensive Systematic Review Methods, Advances and Future Directions

Abstract: Image steganography, which involves the embedding of data into digital images in such a way that the embedding remains undetectable, has seen dramatic changes over the past several years due to advances in deep learning, generative adversarial networks (GANs), diffusion-based generative modelling, and transformer architectures. This paper aims to synthesize recent progress within four main categories of image steganography spatial domain, transform domain, deep-learning-based, and coverless steganography via a review of 40 recent papers published from 2021 to 2025. The methods are reviewed according to a common set of metrics Peak Signal-To-Noise Ratio (PSNR), structural similarity index (SSIM), embedding capacity, invisibility, robustness against steganalysis, and computation cost. Special attention is devoted to four rapidly developing subdomains: diffusion-based generative steganography, invertible neural networks (INNs), adversarially trained hidden mechanisms, and secure methods with provable guarantees. The review additionally includes a structured taxonomy of the field, a comparative analysis based on visual data, a list of popular datasets, an overview of open challenges, and directions for future work.

Keywords: Image steganography, deep learning, generative adversarial networks, diffusion models, invertible neural networks, Coverless steganography, steganalysis, PSNR, SSIM, systematic review.