Detecting and Distinguishing Adaptive and Non-Adaptive Steganography by Image Segmentation

Detecting and Distinguishing Adaptive and Non-Adaptive Steganography by Image Segmentation

Jie Zhu (SKLOIS, Institute of Information Engineering, Chinese Academy of Sciences (CAS), Beijing, China), Xianfeng Zhao (SKLOIS, Institute of Information Engineering, Chinese Academy of Sciences (CAS), Beijing, China) and Qingxiao Guan (SKLOIS, Institute of Information Engineering, Chinese Academy of Sciences (CAS), Beijing, China)
Copyright: © 2019 |Pages: 16
DOI: 10.4018/IJDCF.2019010105
Article PDF Download
Open access articles are freely available for download


This article describes how blind steganalysis aiming at uncovering the existence of hidden data in digital images remains an open problem. Conventional spatial image steganographic algorithms hide data into pixels spreading evenly in the entire cover image, while the content-adaptive algorithms prefer the textural areas and edge regions. In this article, the impact of image content on blind steganalysis is discussed and a practical and extensible approach to distinguish the different types of steganography and construct blind steganalytic detector is proposed. Through the technique of image segmentation, the images are segmented into sub-images with different levels of texture. The classifier only cares for the sub-images which can help modeling the statistical detectability and is trained on sub-images instead of the entire image. Experimental results show the authors' scheme can recognize the type of steganographic methods reliably. The further steps to improve capacity of blind steganalysis based on image segmentation are also mentioned and achieve better performance than ordinary blind steganalysis.
Article Preview

1. Introduction

Steganography and staganalysis have become important topics in the field of information security, and acquired a great deal of attention by researchers all over the world. While steganography embeds the secret to the media file which aims to hide the fact of covert communication, staganalysis tries to find out the embedding modification by the adversary. Digital image files are the usual carriers of secret message due to the simple approach of acquirement and convenient delivery and operation.

The Least significant bit (LSB) is the main embedding channel for digital image files. Two classical but typical embedding ways are LSB replacement and LSB matching which is also called ±1 embedding (Sharp, 2001). LSB methods are commonly used among the many free steganography tools available on the internet. LSB replacement simply replaces the least significant bits of pixel selected to be modified by the message bits after encryption using a secret key (Ker, 2005). This results in the detection of LSB replacement is easy and achieves reliable performance. To improve the ability for resisting attacks, LSB matching is proposed which add or subtract one randomly from the original pixel value when the message bits differ from the least significant bits of pixels. It turns out that LSB matching is harder to attack by some useful detectors in steganalysis of LSB replacement (Ker, 2005). In recent years, adaptive steganography draws lots of attention since it affords high security. The framework of adaptive steganography is utilizing a coding scheme to minimize a well-designed distortion function which models the statistical changes caused by embedding (Vojtěch Holub & Fridrich, 2013). Adaptive Steganography tends to heuristically embed secret data into the complex areas of an image to avoid causing perceptual artifacts and the changes of statistical properties of images content. This also explains why it can outperform the non-adaptive Steganography.

On the other hand, the trend of modern steganalysis is to extract the statistical features and combine with ensemble classifier to construct binary classifier to distinguish cover images and stego images embedded with secret message. Zhao et al. (Zhao, Zhu, & Yu, 2016) summarized systematic and comprehensive definitions of all paradigms of steganalysis, covering not only laboratory research but also the real-world application. In targeted steganalysis it has different principles to detect different types of embedding strategies. For example, the structural analysis is effective to attack the LSB embedding, and features based on selection channel knowledge are applied to analysis up-to-date content-adaptive steganography algorithms (T. D. Denemark, Boroumand, & Fridrich, 2016). Meanwhile, blind steganalysis play a more important role in universal steganalysis. It requires the attacker to design universal features which are suitable for a wide range of steganography algorithm under circumstances of lack of knowledge about the embedding strategy and the embedding payload. As a result, the goal of the statistical features is to uncover the local and global measures sensitive to the pixels or coefficients modification. Although, cover source mismatch is another open problem since the performance of detector degrades dramatically when the distribution of testing images set does not match that of training set (Zhu, Guan, Zhao, Cao, & Chen, 2017).

Complete Article List

Search this Journal:
Volume 14: 1 Issue (2022): Forthcoming, Available for Pre-Order
Volume 13: 6 Issues (2021): 5 Released, 1 Forthcoming
Volume 12: 4 Issues (2020)
Volume 11: 4 Issues (2019)
Volume 10: 4 Issues (2018)
Volume 9: 4 Issues (2017)
Volume 8: 4 Issues (2016)
Volume 7: 4 Issues (2015)
Volume 6: 4 Issues (2014)
Volume 5: 4 Issues (2013)
Volume 4: 4 Issues (2012)
Volume 3: 4 Issues (2011)
Volume 2: 4 Issues (2010)
Volume 1: 4 Issues (2009)
View Complete Journal Contents Listing