Date of Award

2018

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Electrical and Computer Engineering

First Advisor

Jessica Fridrich

Abstract

Today, the most secure steganographic schemes for digital images embed secret messages while minimizing a distortion function that describes the local complexity of the content. Distortion functions are heuristically designed to predict the modeling error, or in other words, how difficult it would be to detect a single change to the original image in any given area. This dissertation investigates how both the design and detection of such content-adaptive schemes can be improved with the use of side-information.

We distinguish two types of side-information, public and private: Public side-information is available to the sender and at least in part also to anybody else who can observe the communication. Content complexity is a typical example of public side-information. While it is commonly used for steganography, it can also be used for detection. In this work, we propose a modification to the rich-model style feature sets in both spatial and JPEG domain to inform such feature sets of the content complexity.

Private side-information is available only to the sender. The previous use of private side-information in steganography was very successful but limited to steganography in JPEG images. Also, the constructions were based on heuristic with little theoretical foundations. This work tries to remedy this deficiency by introducing a scheme that generalizes the previous approach to an arbitrary domain. We also put forward a theoretical investigation of how to incorporate side-information based on a model of images. Third, we propose to use a novel type of side-information in the form of multiple exposures for JPEG steganography.

Share

COinS