The Journal of Science and Engineering Applications (JSEA)

Abstract

Abstract_ Image steganography is an important field in modern information security. It aims to securely transmit a secret message via a digital image and accurately retrieve it upon receipt, by embedding it in the cover image in a secure and  imperceptible manner. Given the limited robustness and capacity of traditional spatial and transformation-based methods, recent research has turned to machine learning-based adaptive embedding strategies, including deep learning, to improve embedding capacity and resistance to steganalysis. In this paper, a survey of representative image steganography methods is presented, With a brief discussion of traditional spatial domain-based and transformation domain-based methods and an in-depth study of machine learning-based techniques such as support vector machines (SVMs), K-means clustering, convolutional neural networks (CNNs), and generative adversarial networks (GANs). The survey includes 40 papers published in peer-reviewed journals from 2017 to 2025 and analyzes the main advantages and disadvantages of the methods referred to for practical purposes. Furthermore, the survey identifies open challenges in imperceptibility, robustness, and security, and indicates potential future research directions for novel adaptive secure machine-learning-based image steganography schemes.