Date of Award
Doctor of Philosophy (PhD)
Engineering and Technology
Dimitrie C. Popescu
Dean J. Krusienski
With the rapid development of digital multimedia technologies, an old method which is called steganography has been sought to be a solution for data hiding applications such as digital watermarking and covert communication. Steganography is the art of secret communication using a cover signal, e.g., video, audio, image etc., whereas the counter-technique, detecting the existence of such as a channel through a statistically trained classifier, is called steganalysis.
The state-of-the art data hiding algorithms utilize features; such as Discrete Cosine Transform (DCT) coefficients, pixel values, motion vectors etc., of the cover signal to convey the message to the receiver side. The goal of embedding algorithm is to maximize the number of bits sent to the decoder side (embedding capacity) with maximum robustness against attacks while keeping the perceptual and statistical distortions (security) low. Data Hiding schemes are characterized by these three conflicting requirements: security against steganalysis, robustness against channel associated and/or intentional distortions, and the capacity in terms of the embedded payload. Depending upon the application it is the designer's task to find an optimum solution amongst them.
The goal of this thesis is to develop a novel data hiding scheme to establish a covert channel satisfying statistical and perceptual invisibility with moderate rate capacity and robustness to combat steganalysis based detection. The idea behind the proposed method is the alteration of Video Object (VO) trajectory coordinates to convey the message to the receiver side by perturbing the centroid coordinates of the VO. Firstly, the VO is selected by the user and tracked through the frames by using a simple region based search strategy and morphological operations. After the trajectory coordinates are obtained, the perturbation of the coordinates implemented through the usage of a non-linear embedding function, such as a polar quantizer where both the magnitude and phase of the motion is used. However, the perturbations made to the motion magnitude and phase were kept small to preserve the semantic meaning of the object motion trajectory.
The proposed method is well suited to the video sequences in which VOs have smooth motion trajectories. Examples of these types could be found in sports videos in which the ball is the focus of attention and exhibits various motion types, e.g., rolling on the ground, flying in the air, being possessed by a player, etc. Different sports video sequences have been tested by using the proposed method. Through the experimental results, it is shown that the proposed method achieved the goal of both statistical and perceptual invisibility with moderate rate embedding capacity under AWGN channel with varying noise variances. This achievement is important as the first step for both active and passive steganalysis is the detection of the existence of covert channel.
This work has multiple contributions in the field of data hiding. Firstly, it is the first example of a data hiding method in which the trajectory of a VO is used. Secondly, this work has contributed towards improving steganographic security by providing new features: the coordinate location and semantic meaning of the object.
"Data Hiding in Digital Video"
(2011). Doctor of Philosophy (PhD), dissertation, Engineering and Technology, Old Dominion University, DOI: 10.25777/kmnq-nd55