Robust and Blind 3D Watermarking

3D watermarking is a technique to hide some information into the 3D graphical
model in such a way that the watermarked object is visually indistinguishable from
the original one. A robust and blind 3D watermarking method should be able to
detect the embedded message after a certain level of malicious attack without having
the original model. 3D watermarking has a great potential of usage in the real world
and it can be applied in the copyright protection, database management, graphics
authentication and data transmission etc. This thesis proposes four novel robust
and blind 3D watermarking methods based on spectral domain and spatial domain.
Chapter 2 comprehensively surveys the related literature in the fields of transformed
domain methods, spatial domain methods and the watermarking metrics.
Chapter 3 proposes a novel 3D watermarking methodology in the spectral domain.
The mesh object is decomposed into a set of spectral coefficients which represent
the energy of the mesh in different scales. The message is embedded by introducing
constraints into the distributions of spectral coefficients.
Chapter 4 employs the geodesic distance to carry the bits based on the observation
that the distribution of geodesic distance within a range is close to uniform.
Two ways of embedding scheme are introduced. One is to modify the mean value and
the distribution and the other is to change the variance. A novel Vertex Placement
Scheme (VPS) is proposed to move the vertex in order to satisfy the watermarked
geodesic distance, without causing significant distortion to the object.
Chapter 5 introduces two spatial domain methods which embed the message by
changing the distribution of the vertex norms, i.e. the distance from vertex to the
object centre. Two methods employ the same histogram mapping function as described
in chapter 4. The first method minimizes the surface distortion by selecting
a candidate point over the neighbourhood which introduces the minimum error. The
second method employs the Levenberg-Marquardt optimization method to find the
best possible solution to ensure that the surface distortion is truly minimum with
respect to a novel surface error function.
The algorithms proposed in this thesis significantly improve the visual quality of
the watermarked object while the watermark detection robustness is at a relatively
high level. The robustness of the proposed methods is increasing from the methods
presented in Chapter 3 to Chapter 5 while the surface distortion is decreasing for
these methods. The second algorithm proposed in Chapter 5 achieves the best
overall performance in the aspect of visual quality and robustness. In Chapter 6,
we conclude the thesis by addressing the weakness and propose potential future
research work.