Steganography is a secret communication method in which the actual message is hidden in some other innocuous looking object. Digital media, such as images, are ideal for this purpose because they can be slightly modified to encode a secret without making these modifications visible to a human being or detectable by a computer. Steganography thus offers privacy to citizens in countries that censor communication channels or prohibit the usage of encryption. This project studies the fundamental principles and limits of such stealth communication by starting with a detailed mathematical understanding of how a digital image is formed inside the camera. This provides mathematical guarantee that the hidden data cannot be discovered by an adversary, protecting thus the users. The model is also expected to find applications in a related field of digital forensics that tries to identify which portion of a digital image has been altered, what kind of camera took the image, and what type of enhancement it was subjected to. This project is closely related to homeland security, intrusion detection and its prevention, information assurance, and trusted information exchange.
Steganography in digital media has historically relied on heuristic reasoning and empirical evidence of security due to the complexity associated with the cover medium and the lack of tractable models. This makes it difficult, if possible, to establish optimality, establish security formally, and assess threats. The main thrust of this project is to move from heuristics to approaches that are based on tight-fitting models to construct provably secure embedding schemes and assess their security formally. The investigation begins with a domain in which true randomness with a tractable mathematical description exists, which is then propagated to the representation in which the actual data hiding occurs. This will be achieved by adopting pixel-specific models for the RAW sensor capture and deriving a model in the embedding domain in a closed- form, if the complexity of the development (processing) allows, or by deriving (estimating) a tightly-fitting model using Gaussian Markov random fields, parametric multivariate models, Monte-Carlo sampling, and using data-driven models. Multiple closely related tasks will be investigated that involve reasoning based on statistical hypothesis testing and information theory to provide the foundation for design, assessment, and analysis of steganographic systems, and to establish capacity of covert channels at a given level of statistical detectability. The project has the potential to improve theoretical understanding of the complex interaction between the users of covert tools and the adversaries in terms of payload scaling laws, bounds on secure covert communication rates, and on statistical detectability of the covert communication channel. Leveraging tightly fitting models of indeterministic components within digital media will help remove heuristics and intuition so often relied upon in digital-media steganography. Pixel-specific sensor capture models will find applications in the related field of digital media forensics for determining provenance, origin, and integrity.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
