AbstractUnlike other security methods, steganography hides the very existence of secret messages rather than their content only. Both steganography and steganalysis are strongly related to each other, the new steganographic methods should be evaluated with current steganalysis methods and vice-versa. Since steganography is considered broken when the stego object is recognised, undetectability would be the most important property of any steganographic system. Digital image files are excellent media for steganography, as they have redundancy in their representation. Also, the most widely used method of image steganography is the least significant bit (LSB) embedding.
This thesis investigates the latest methods of pixel domain steganography and provides new efficient approaches to improve them in three perspectives: embedding, detection, and the digital forensics investigation process. Firstly, the probability of detection is considered for non-adaptive LSB and 2LSB image steganography even for the embedding rate of 1. The proposed method noticeably reduced the probability of detection for different detection methods via improving the embedding efficiency of both LSB and 2LSB methods, which is not restricted to a specific steganalysis attack.
The extensions to LSB steganography methods have received great attention from steganographers, especially 2LSB, because it is easy to implement, has a higher capacity, is visually imperceptible, brings complex changes to the image pixel values and is harder to detect. The proposed method improves the detection accuracy of the current state of the art targeted 2LSB steganalysis methods via a novel approach pixel value grouping and statistical analysis of the image pixel values histogram. Moreover, a discrete classifier version of the proposed method is developed which gives a label (‘Stego’ or ‘Clean’) to the analysed image and avoids the overhead of setting a right threshold value.
The last perspective of this research considers the evaluation process of the steganalysis tools and simplifying the digital forensics investigation process. Hence, a novel statistical method is proposed to effectively simplify the investigation process by showing the area of differences between the testing image set and the random set of images that is used as a baseline. It also indicates whether the difference is significant or not.
All the above mentioned novel approaches included in this thesis are proven, in both theoretical and practical perspectives, to be better than the current state-of-the-art methods and add some value to the knowledge in the field of steganography, steganalysis and its applications.
Key words: Steganography, Steganalysis, LSB embedding, 2LSB embedding, Forensic steganalysis, LSB embedding, 2LSB steganalysis
|Date of Award||Dec 2015|
|Supervisor||Benjamin Aziz (Supervisor), Carl Adams (Supervisor) & Julio Hernandez-Castro (Supervisor)|