Article Preview
TopIntroduction
Many three-dimensional (3D) objects are now represented in 3D meshes to truly reflect the topological structures of the objects. Among various representation tools, triangular meshes provide an effective means to represent 3D mesh models. With high demand and popularity of 3D models and considering the cost, time, and effort required to build such models comes the menace of widespread illegal copying of 3D models. Watermarking is a technique which deters illegal copying by inserting a hidden message in the 3D model (Matwini, 2011). Digital watermarking does not restrict access to the host data, but ensures the hidden data to remain inviolate and recoverable. Watermarking is a copyright protection technique to embed information, so-called watermark, into host data (Kim; Valette; Jung; & Prost, 2005).
According to the aimed application, we distinguish between the robust watermark used for intellectual property protection and the fragile watermark used for content authentication (Wanga, Lavou´ea, Denisb, & Baskurta, 2010) . A Robust watermarks are designed to resist attempts to remove or destroy the watermark. Their primary applications are copyright protection and content tracking. Unlike robust watermarks, fragile watermarks are designed to be easily destroyed if the watermarked model is manipulated in the slightest manner. This property is ideal for authentication applications (Liu & Qiu, 2002).
Depending on whether the original cover-media is needed or not in the detection stage we have non-blind and blind watermarking. The methods from the first category usually have good robustness, but they are not suitable for most applications (Soliman, Abo, & Onsi, 2013).
Since the watermarking technique for 3D meshes was introduced by (Ohbuchi, Masuda, & Aono, 1997) there have been several trials to improve the performance in terms of capacity, invisibility and robustness. Most of the existing methods concern mesh models and can be classified into two main categories, depending if the watermark is embedded in the spatial or in the spectral domain. Spatial domain based techniques are quite fast and simple to implement, but do not yet provide enough robustness and are rather adapted for blind fragile watermarking or steganography (Lavou´e, Denis, Dupont, & Baskurt, 2006). These set of spatial domain technique consist of embedding the watermark information directly by modifying either the 3D mesh geometry or the topology of the triangles (Ohbuchi, 1998; Benedens, 1999; Yeo, 1999; Cayre, 2003).
The other category is based on the transformation domain. In these algorithms as introduced in (Valette, 2004; Praun, 1999; Kanai, 1998; Yin, 2001; Ohbuchi, 2001; Ohbuchi, 2002),spectral decomposition and multi resolution techniques such as wavelet transform and progressive meshes are used to decompose a 3D model into a lower resolution and the watermark is inserted in the bit stream, the model is then reconstructed from lower resolutions (Soliman, Abo, & Onsi, 2013). While in the frequency domain watermarking method, the watermarking process is more complex and slower than the spatial one because of the need to transform and inverse transform. However, the embedded watermark is stronger to the operations (Kanai, Date, & Kishinami, 1998).
This paper introduces a novel approach that utilizes the artificial intelligence technique in optimally selecting the locations where the watermark needs to be inserted by exploiting the local geometry of the 3D model. This approach introduces a robust and blind 3D watermarking approach using clustering techniques such as k-means and Self Organizing Map (SOM). Both techniques are used to cluster mesh vertices into suitable and non-suitable watermark carrier based on local geometry measures. Once the vertices are clustered to be suitable carriers for watermark bit stream, the insertion of watermark bits is performed based on local statistical measures such as mean and stander deviation.
Organization of the paper is as follows. Some useful and important preliminary ideas relating to this work are presented in the following section. The proposed approach is outlined later followed by the experimental results and conclusions.