Information Analysis for Steganography and Steganalysis in 3D Polygonal Meshes

Abstract

Information hiding, which embeds a watermark/message over a cover signal, has recently found extensive applications in, for example, copyright protection, content authentication and covert communication. It has been widely considered as an appealing technology to complement conventional cryptographic processes in the field of multimedia security by embedding information into the signal being protected. Generally, information hiding can be classified into two categories: steganography and watermarking. While steganography attempts to embed as much information as possible into a cover signal, watermarking tries to emphasize the robustness of the embedded information at the expense of embedding capacity.

In contrast to information hiding, steganalysis aims at detecting whether a given medium has hidden message in it, and, if possible, recover that hidden message. It can be used to measure the security performance of information hiding techniques, meaning a steganalysis resistant steganographic/watermarking method should be imperceptible not only to Human Vision Systems (HVS), but also to intelligent analysis.

As yet, 3D information hiding and steganalysis has received relatively less attention compared to image information hiding, despite the proliferation of 3D computer graphics models which are fairly promising information carriers. This thesis focuses on this relatively neglected research area and has the following primary objectives: 1) to investigate the trade-off between embedding capacity and distortion by considering the correlation between spatial and normal/curvature noise in triangle meshes; 2) to design satisfactory 3D steganographic algorithms, taking into account this trade-off; 3) to design robust 3D watermarking algorithms; 4) to propose a steganalysis framework for detecting the existence of the hidden information in 3D models and introduce a universal 3D steganalytic method under this framework. %and demonstrate the performance of the proposed steganalysis by testing it against six well-known 3D steganographic/watermarking methods.

The thesis is organized as follows. Chapter 1 describes in detail the background relating to information hiding and steganalysis, as well as the research problems this thesis will be studying. Chapter 2 conducts a survey on the previous information hiding techniques for digital images, 3D models and other medium and also on image steganalysis algorithms.

Motivated by the observation that the knowledge of the spatial accuracy of the mesh vertices does not easily translate into information related to the accuracy of other visually important mesh attributes such as normals, Chapters 3 and 4 investigate the impact of modifying vertex coordinates of 3D triangle models on the mesh normals. Chapter 3 presents the results of an empirical investigation, whereas Chapter 4 presents the results of a theoretical study. Based on these results, a high-capacity 3D steganographic algorithm capable of controlling embedding distortion is also presented in Chapter 4.

In addition to normal information, several mesh interrogation, processing and rendering algorithms make direct or indirect use of curvature information. Motivated by this, Chapter 5 studies the relation between Discrete Gaussian Curvature (DGC) degradation and vertex coordinate modifications.

Chapter 6 proposes a robust watermarking algorithm for 3D polygonal models, based on modifying the histogram of the distances from the model vertices to a point in 3D space. That point is determined by applying Principal Component Analysis (PCA) to the cover model. The use of PCA makes the watermarking method robust against common 3D operations, such as rotation, translation and vertex reordering. In addition, Chapter 6 develops a 3D specific steganalytic algorithm to detect the existence of the hidden messages embedded by one well-known watermarking method. By contrast, the focus of Chapter 7 will be on developing a 3D watermarking algorithm that is resistant to mesh editing or deformation attacks that change the global shape of the mesh.

By adopting a framework which has been successfully developed for image steganalysis, Chapter 8 designs a 3D steganalysis method to detect the existence of messages hidden in 3D models with existing steganographic and watermarking algorithms. The efficiency of this steganalytic algorithm has been evaluated on five state-of-the-art 3D watermarking/steganographic methods. Moreover, being a universal steganalytic algorithm can be used as a benchmark for measuring the anti-steganalysis performance of other existing and most importantly future watermarking/steganographic algorithms.

Chapter 9 concludes this thesis and also suggests some potential directions for future work.