| Summary: | 博士 === 國立中興大學 === 資訊科學系所 === 95 === Throughout history, people have sought secret means of protecting their property and their knowledge from being stolen. Watermarking and copyright marking, for instance, attempt to keep property secure from thieves. Cryptography and steganography aim to keep knowledge safe from spies. Whereas cryptography seeks to disguise the content of messages, steganography is the art and science of communicating in a way which hides the existence of the communication itself. Although the term was only coined at the very end of the fifteenth century, the use of steganography dates back several millennia. In antiquity, it saved many empires when it was used to transmit undetectable secret information. Unlike copyright marking and cryptography, steganography conceals messages inside other harmless media in a way that does not allow any interceptors to even detect that there is a second, secret message present. When combined with the other techniques for guarding property and message content, steganography can achieve a higher level of security.
As an increasing amount of data is stored on computers, it is no surprise that steganography has entered the digital age. The Internet is such a popular and convenient communication channel that more and more digital media are being transmitted over networks. Protecting secret messages during transmission is thus an important issue in the development of Internet technologies. In computer-based steganography, several forms of digital media may be used as cover for hidden information, including three-dimensional (3D) models and high dynamic range (HDR) images. The recent increased use and importance of such models and images have prompted further investigations of steganographic techniques for the protection of secret messages. Since these models include 3D polygonal meshes and point-sampled geometries, this dissertation proposes several advanced techniques to solve some of the most pressing steganographic problems for these modals and HDR images.
Even though information hiding is a very active research field and its application to traditional cover media, such as 2D still images, has been rather thoroughly studied, information hiding using 3D models has not been heavily researched. Several steganographic schemes have been presented for 3D models, but most of them provide poor capacity and entail some visual distortion. Moreover, processing is always time consuming, and most of the approaches are only dedicated to a particular model type, such as 3D triangular models or point-sampled geometries, because the representation of such models and such geometries is very different.
As a result, there has been an urgent need to develop a more powerful steganographic algorithm for 3D models. Ideally, this algorithm should fully exploit the features of 3D models based on the consideration of 3D space to efficiently embed a large number of messages into a 3D model with insignificant model distortion. It would also be a significant improvement in that it could support both 3D polygonal meshes and point-sampled geometries simultaneously. We have met this need by designing just such an effective steganographic algorithm, as each chapter in this dissertation illustrates.
We propose several efficient traversal techniques for deriving the embedding sequence list for 3D models to improve the performance. We first present an efficient triangle mesh traversal technique for 3D triangular models. Then, inspired by the concept of epidemics, we introduce a “contagious diffusion technique” for general polygonal meshes, extending its feasibility to other non-triangular mesh applications. This new technique is simpler and more efficient than any other method currently available. Finally, we also offer an innovative abscissa projection technique, which decides the embedding sequence list for point-sampled geometries by employing principal component analysis (PCA).
From the point of view of capacity and visual appearance, our efficient and high-capacity steganographic techniques for 3D models significantly decrease the visual degradation normally produced by similar techniques; we do this by successfully exploring the characteristics of the human visual system (HVS). Moreover, we accomplish this in two key steganographic areas: the spatial domain and the representation domain.
In the spatial domain, we first present an efficient multi-level embedding procedure for 3D polygonal meshes. Our algorithm, which is the first technique to take advantage of the full vertex features of these meshes, achieves high capacity relying on three independent degrees of freedom. We deal with distortion through a new modified multi-level embedding procedure. Embedding that relies on the angle between triangle planes is not appropriate to a long and narrow triangle, because such embedding can cause larger distortion for a larger radius. To solve this, we embed messages based on the arc length, which efficiently avoids apparent distortion while it achieves high capacity. Based on this approach, we also propose a virtual multi-level embedding procedure to support point-sampled geometries. It produces a unique virtual triangle for each point and embeds messages by modifying the point based on virtual geometrical properties. Then, we propose an adaptive minimum-distortion estimation procedure to achieve higher capacity in each vertex by changing the vertex’s position with minimum distortion distance. The proposed approach considers the correlation between a polygon and its neighbor polygons with respect to human visual perception. To the best of our knowledge, it is the first adaptive 3D polygonal mesh steganographic scheme that can achieve adaptability and preserve important shape features, such as ridges and corners.
Our techniques are also successful in the representation domain. Based on the research of embedding messages into a 3D model with insignificant distortion, we present a novel distortion-free steganographic algorithm, including a representation rearrangement procedure and an adaptive representation rearrangement procedure, for 3D models. Our algorithm can support both 3D polygonal meshes and point-sampled geometries. To the best of our knowledge, this is the first steganographic approach for 3D models that can utilize representation information to achieve high capacity without introducing any visual distortion.
As we demonstrate in this dissertation, the proposed approaches, based on both the spatial and representation domains, can be combined to achieve higher capacity. In other words, we successfully combine both domains for steganography. Experimental results show significant improvements in terms of capacity, visual appearance, and performance with respect to the most recent, advanced techniques. These improvements will make this kind of application more widely available.
Finally, all of our approaches are blind schemes, requiring no 3D cover model for message extraction, and they are secure, since retrieving the message without the key is virtually impossible. These algorithms are simple yet efficient and are robust against affine transformations. Our techniques are a feasible alternative to steganographic approaches for 3D models.
Besides the discussion of the steganographic approaches for 3D models, we also propose a specialized steganographic approach for HDR images. In recent years, there has been an explosion of interest and need in HDR images, which had lead to HDR images being increasingly popular in various fields. However, research in steganography has not kept pace with the advances of HDR images, even though numerous steganography schemes have been presented for conventional low dynamic range (LDR) images. Developing a steganographic algorithm for HDR images presents a distinct challenge because, unlike the fixed range of luminance for LDR image, each HDR image has a very different luminance range. Such an algorithm must be general enough to cope with very different ranges of HDR images. To the best of our knowledge, our scheme is the first completely feasible steganographic approach for HDR images. To achieve adaptive message embedding, we exploit the correlation of color, luminance, and contrast between neighboring pixels in accordance with normal human visual perception. In addition, our scheme also provides the capability of authentication, which can detect whether a stego HDR image has been tampered with before the message extraction. Experimental results show that our scheme has large capacity with unnoticeable distortion. Our technique is adaptive, simple, efficient, and secure, and has proven to be feasible in steganography.
In short, this dissertation presents the main issues related to the context of steganography for 3D models. We discuss steganography for HDR images, and we explore the main applications making use of secret messages and their need of reliable, high-capacity techniques. Among these techniques, we restrict our work to blind steganographic schemes with high capacity, adaptability, and no distortion. These techniques are simple, efficient, and generalizable. Moreover, they are secure in the cryptographic sense.
Because research in this field is growing and diversifying, a great deal of additional work is needed. Development in the area of steganography will continue. The novel and original ideas for steganographic applications we offer in this dissertation can benefit many applications in cryptography and copyright marking, as well. We hope that our work will inspire more research in these areas.
|
|---|
