VIDEO STEGANOGRAPHY TECHNIQUES: A COMPARATIVE STUDY OF SPATIAL, TRANSFORM-DOMAIN, CODEC-AWARE AND DEEP NEURAL LEARNING APPROACHES

Abstract

Video steganography conceals secret information within digital video while attempting to preserve visual quality, extraction reliability and resistance to statistical detection or signal-processing attacks. The field has progressed from direct least-significant-bit substitution to transform-domain embedding, motion-vector and codec-element manipulation, region-of-interest selection, optimization-based schemes, robust social-media transmission and deep neural video hiding. This paper presents a structured comparative review of representative approaches published from foundational work through 2025. The methods are analyzed using a common framework comprising embedding domain, payload capacity, imperceptibility, robustness, security against steganalysis, computational complexity, codec dependence and practical deployment. Particular attention is given to the randomized frame selection method which combines a 64-bit secret-key-derived frame index, payload encryption and LSB embedding. The method reports very high visual fidelity in its experiments, including a maximum PSNR of 74.15 dB and an MSE of 0.0002, while improving location secrecy over fixed-frame LSB schemes. However, its robustness against transcoding, frame deletion, resizing and modern learning-based steganalysis remains insufficiently established. The comparative synthesis shows that spatial methods offer high capacity and low cost but weak robustness; transform and codec-aware methods improve resilience at the expense of complexity and codec dependence; and neural methods achieve flexible high-capacity hiding but require training resources and more rigorous security evaluation. The paper concludes that future systems should combine key-controlled temporal selection, content-adaptive embedding, codec-aware robustness, error correction and steganalysis-aware learning under standardized evaluation protocols.