Abstract

Multimedia steganography is the method of hiding information within images, audio, video, and text. The ability to detect hidden data in multimedia steganography is important for cybersecurity and secure communication systems. Historically, steganalysis techniques were conducted manually and proved to be less effective against sophisticated hiding techniques currently developed using deep neural networks and their hybrid models. Recent improvements in deep convolutional neural networks and transformers have led to better detection of local and global patterns. In this review, we provide a comprehensive survey of recent developments in advanced multimedia steganalysis, particularly highlighting the hybrid approaches combining CNNs and Transformers, as well as CNNs and LSTMs. This review classifies the current steganalysis techniques based on existing studies, evaluates their performance on widely used benchmarks, examines architectural designs, and summarizes the latest advancements related to image, audio, video, and text-based steganalysis. This review highlights the need for new approaches like cross-media generalized steganalysis, multi-scale fusion methods, attention models, and LLM-based semantic steganalysis.