Quantitative and Binary Steganalysis in JPEG: A Comparative Study

Ahmad Zakaria (1) , Marc Chaumont (1) , Gérard Subsol (1)

1 ICAR - Image & Interaction

We consider the problem of steganalysis, in which Eve (the steganalyst) aims to identify a steganogra-pher, Alice who sends images through a network. We can also hypothesise that Eve does not know how many bits Alice embed in an image. In this paper, we investigate two different steganalysis scenarios: Binary Steganalysis and Quantitative Steganalysis. We compare two classical steganalysis algorithms from the state-of-the-art: the QS algorithm and the GLRT-Ensemble Classifier, with features extracted from JPEG images obtained from BOSSbase 1.01. As their outputs are different, we propose a methodology to compare them. Numerical results with a state-of-the-art Content Adaptive Embedding Scheme and a Rich Model show that the approach of the GLRT-ensemble is better than the QS approach when doing Binary Steganalysis but worse when doing Quantitative Steganalysis.

Mots clés: Multi-class Steganalysis; Binary Steganalysis; JPEG; Steganography; Quantitative Steganalysis

Domaines: Cryptographie et sécurité [cs.CR]; Apprentissage [cs.LG]; Traitement du signal et de l'image [eess.SP]

Ahmad Zakaria, Marc Chaumont, Gérard Subsol. Quantitative and Binary Steganalysis in JPEG: A Comparative Study. EUSIPCO: European Signal Processing Conference, Sep 2018, Rome, Italy. pp.1422-1426. ⟨lirmm-01884006⟩