Exploiting similarities between secret and cover images for improved embedding efficiency and security in digital steganography

Abdulla, Alan Anwer (2015) Exploiting similarities between secret and cover images for improved embedding efficiency and security in digital steganography. Doctoral thesis, University of Buckingham.

Alan Abdulla - PhD Thesis.pdf
Available under License Creative Commons Attribution Non-commercial No Derivatives.

Download (3MB) | Preview


The rapid advancements in digital communication technology and huge increase in computer power have generated an exponential growth in the use of the Internet for various commercial, governmental and social interactions that involve transmission of a variety of complex data and multimedia objects. Securing the content of sensitive as well as personal transactions over open networks while ensuring the privacy of information has become essential but increasingly challenging. Therefore, information and multimedia security research area attracts more and more interest, and its scope of applications expands significantly. Communication security mechanisms have been investigated and developed to protect information privacy with Encryption and Steganography providing the two most obvious solutions. Encrypting a secret message transforms it to a noise-like data which is observable but meaningless, while Steganography conceals the very existence of secret information by hiding in mundane communication that does not attract unwelcome snooping. Digital steganography is concerned with using images, videos and audio signals as cover objects for hiding secret bit-streams. Suitability of media files for such purposes is due to the high degree of redundancy as well as being the most widely exchanged digital data. Over the last two decades, there has been a plethora of research that aim to develop new hiding schemes to overcome the variety of challenges relating to imperceptibility of the hidden secrets, payload capacity, efficiency of embedding and robustness against steganalysis attacks. Most existing techniques treat secrets as random bit-streams even when dealing with non-random signals such as images that may add to the toughness of the challenges.This thesis is devoted to investigate and develop steganography schemes for embedding secret images in image files. While many existing schemes have been developed to perform well with respect to one or more of the above objectives, we aim to achieve optimal performance in terms of all these objectives. We shall only be concerned with embedding secret images in the spatial domain of cover images. The main difficulty in addressing the different challenges stems from the fact that the act of embedding results in changing cover image pixel values that cannot be avoided, although these changes may not be easy to detect by the human eye. These pixel changes is a consequence of dissimilarity between the cover LSB plane and the secretimage bit-stream, and result in changes to the statistical parameters of stego-image bit-planes as well as to local image features. Steganalysis tools exploit these effects to model targeted as well as blind attacks. These challenges are usually dealt with by randomising the changes to the LSB, using different/multiple bit-planes to embed one or more secret bits using elaborate schemes, or embedding in certain regions that are noise-tolerant. Our innovative approach to deal with these challenges is first to develop some image procedures and models that result in increasing similarity between the cover image LSB plane and the secret image bit-stream. This will be achieved in two novel steps involving manipulation of both the secret image and the cover image, prior to embedding, that result a higher 0:1 ratio in both the secret bit-stream and the cover pixels‘ LSB plane. For the secret images, we exploit the fact that image pixel values are in general neither uniformly distributed, as is the case of random secrets, nor spatially stationary. We shall develop three secret image pre-processing algorithms to transform the secret image bit-stream for increased 0:1 ratio. Two of these are similar, but one in the spatial domain and the other in the Wavelet domain. In both cases, the most frequent pixels are mapped onto bytes with more 0s. The third method, process blocks by subtracting their means from their pixel values and hence reducing the require number of bits to represent these blocks. In other words, this third algorithm also reduces the length of the secret image bit-stream without loss of information. We shall demonstrate that these algorithms yield a significant increase in the secret image bit-stream 0:1 ratio, the one that based on the Wavelet domain is the best-performing with 80% ratio.For the cover images, we exploit the fact that pixel value decomposition schemes, based on Fibonacci or other defining sequences that differ from the usual binary scheme, expand the number of bit-planes and thereby may help increase the 0:1 ratio in cover image LSB plane. We investigate some such existing techniques and demonstrate that these schemes indeed lead to increased 0:1 ratio in the corresponding cover image LSB plane. We also develop a new extension of the binary decomposition scheme that is the best-performing one with 77% ratio. We exploit the above two steps strategy to propose a bit-plane(s) mapping embedding technique, instead of bit-plane(s) replacement to make each cover pixel usable for secret embedding. This is motivated by the observation that non-binary pixel decomposition schemes also result in decreasing the number of possible patterns for the three first bit-planes to 4 or 5 instead of 8. We shall demonstrate that the combination of the mapping-based embedding scheme and the two steps strategy produces stego-images that have minimal distortion, i.e. reducing the number of the cover pixels changes after message embedding and increasing embedding efficiency. We shall also demonstrate that these schemes result in reasonable stego-image quality and are robust against all the targeted steganalysis tools but not against the blind SRM tool. We shall finally identify possible future work to achieve robustness against SRM at some payload rates and further improve stego-image quality.

Item Type: Thesis (Doctoral)
Uncontrolled Keywords: Digital steganography; steganalysis; hidden images; image files
Subjects: Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Depositing User: Users 4 not found.
Date Deposited: 29 Sep 2016 08:43
Last Modified: 12 Dec 2019 14:55
URI: http://bear.buckingham.ac.uk/id/eprint/149

Actions (login required)

View Item View Item