MASTER THESIS Video Watermarking - Computer Graphics Group ...
MASTER THESIS Video Watermarking - Computer Graphics Group ...
MASTER THESIS Video Watermarking - Computer Graphics Group ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
But the description is restricted to the former values in order to be less confusing;<br />
proposed algorithms, processes and calculations do not change.<br />
Figure 10 illustrates content ID spreading, watermark generation and<br />
embedding which is described below.<br />
<br />
<br />
<br />
<br />
<br />
<br />
Figure 10: Illustration of watermark generation and embedding<br />
Once the watermark is generated and hidden information is encoded, the<br />
embedding process can take place.<br />
Blocks of each macroblock could be watermarked using formula (2) in case of<br />
frequency domain watermarks (the transform coefficients of a block are altered to<br />
encode one watermark element) or formula (4) in case of spatial domain<br />
watermark (corresponding sub-block of watermark block elements is forward<br />
transformed and added to the coefficients).<br />
However, it not as simple as it seems. The essence of the problem consists in<br />
intra prediction. If watermarked blocks are used for intra prediction of other<br />
blocks, distortion caused by watermark embedding spreads into the other blocks.<br />
If there is a sequence of predictively dependent blocks, the distortion propagates<br />
and accumulates up into the last block of the sequence which probably causes<br />
severe, obviously not unobtrusive, fidelity distortion. Therefore, the intra<br />
prediction error compensation is implemented to undo the distortion.<br />
In one block of a macroblock, the embedding process proceeds as follows (the<br />
scheme is depicted in Figure 11). The residual is obtained using inverse<br />
frequency transform on dequantized transform coefficients. Then, the predicted<br />
samples of both the original and the watermarked pictures are computed. Thus,<br />
both pictures are constructed during the process. The residual is added to<br />
predicted samples of the original picture, clipped to allowed range and stored as a<br />
block of the original picture. The prediction error as the difference between the<br />
watermarked picture predicted samples and the original picture predicted<br />
24