Steganalysis of content-adaptive steganography in spatial domain

Abstract

Content-adaptive steganography constrains its embedding changes to those parts of covers that are difficult to model, such as textured or noisy regions. When combined with advanced coding techniques, adaptive steganographic methods can embed rather large payloads with low statistical detectability at least when measured using feature-based steganalyzers trained on a given cover source. The recently proposed steganographic algorithm HUGO is an example of this approach. The goal of this paper is to subject this newly proposed algorithm to analysis, identify features capable of detecting payload embedded using such schemes and obtain a better picture regarding the benefit of adaptive steganography with public selection channels. This work describes the technical details of our attack on HUGO as part of the BOSS challenge. © 2011 Springer-Verlag.