Research On Stereoscopic Image Digital Watermarking Technology Based On Visual Perception

With the viewer can perceive depth while looking at a stereoscopic screen, three dimensionalmedia is becoming the next important developing direction. Many movies are already recorded in astereoscopic format today.3D video has faced a new set of challenging problems such as copyrightprotection, authentication and content integrity. Digital watermarking is a process of embeddingunobtrusive marks or labels into digital content. These embedded marks are typically imperceptiblethat can later be detected or extracted. There is also a strong necessity for developing stereoscopicimage watermark techniques, which can solve these problems. And effective digital watermark forstereoscopic image will play an important role for promoting the entertainment industry.It is necessary to protect the stereoscopic images’ ownership firstly in order to solve thecopyright problems for3D video. Presently, digital watermarking has been employed as a meansfor indicating the ownership of copyrighted multimedia data. Digital watermarking has beenextensively studied for two dimensional images, but very few algorithms have been proposed forthe watermarking of stereo images. This thesis described four watermarking algorithms forstereoscopic images. They are concluded as followed:While many existing visual sensitivity models for images did not consider the depth ordisparity of stereoscopic images, a visual sensitivity model is proposed for watermarking ofstereoscopic images. Firstly, visual sensitivity models in pixel and frequency domain are described.After that, taking depth and disparity into consideration, an effective visual sensitivity model isdiscussed to guide stereoscopic image watermarking scheme in the DCT domain.Secondly, many existing watermarking methods did not take visual sensitivity and thefeatures of stereoscopic images into consideration. Three watermarking algorithms are proposed,they are discussed as followed:1) A robust algorithm is proposed by making use of the stereoscopic images’ features in thischapter. Firstly, global disparity curve is used to describe the stereoscopic images’ features andsaved as a key. Watermark is embedded in similar areas selected by global disparity. Every bit ofwatermark is embedded in similar blocks by exploiting relationship embedding. In order to achievethe best trade-off point of robustness and transparency of the watermarked images, thewatermarking scheme is based on the Watson visual sensitivity model to determine the perceptualadjustment on watermark insertion. Similar blocks usually have the same change trend when twoviewpoints are under the same attack. The result of this experiment shows that the bit same rate ofthe extracted watermark is around0.85. The proposed method is effective to a kind of attacks.2) A relationship-based watermarking algorithm is proposed to protect copyright ofstereoscopic images. The biggest difference between stereoscopic images and two-dimensional pictures is that there are two views in stereoscopic images. The proposed algorithm is not onlyembedding watermark in left image, but also embedding watermark in right image by quantifyingthe differences between discrete wavelet coefficients of watermarked left image and right image.The step is most important to adjust the tradeoff of robustness and transparency of the watermarkedimages. A lot of experiments are performed to determine the final step. Experimental results showthe bit same rate is higher than0.85, and more robust than the former one.3) Matching-based watermarking algorithm is proposed by making use of the similar matchedblocks of stereoscopic images. In this section, a method for improving the robustness of watermarkby considering visual sensitivity model of stereoscopic images. Experimental results show that thewatermark image in stereoscopic image is invisible. And the algorithm is much more robust thanthe others.Thirdly, a HVS-based multipurpose watermarking algorithm is proposed，which can achievethe goal of content authentication and copyright protection of stereoscopic images simultaneously.More than one watermark is embedded into the same multimedia object in multipurposewatermarking algorithm. There is a problem that the latter embedding watermark has an effect tothe former one, making it harder to guarantee the quality of the watermarks. In the proposedmethod, the semi-fragile watermark and the robust watermark are embedded in different regions.Semi-fragile watermark is embedded in region of interest (ROI) to detect the stereoscopic imagesis under malicious attacks or not. Robust watermark is embedded in region of non-interest toprotect copyright of stereoscopic images. Experimental results demonstrate the effectiveness of thealgorithm in terms of robustness and semi-fragility.