Graphics Reference
In-Depth Information
4
Experimental Results
In our experiments, various attacks are tested to evaluate the robustness of our method,
including amplitude scaling, image filtering, adding noise, rotation and resizing. All the
test images are of size 512
512 and in gray-scale. All test images are decomposed with
three level wavelets, and the block size w is 4. The peak signal-to-noise ratio (PSNR)
is used to measure the similarity of the original image to the watermarked image. And
the bit error rate (BER) is used to judge the existence of the watermark.
×
4.1
Comparison with Basic RDM
In this section, we compare the improved RDM algorithm with the basic RDM method.
We denote the basic RDM algorithm as RDM-Basic. The RDM-Basic is implemented
with 10th, 30th and 50th order respectively. The watermark is a 1024 bits Gaussian
pseudo-random sequence. To fill up the capacity, the watermark is embedded with four
times repeatedly in RDM-Basic method, while in the proposed improved RDM method,
the watermark is embedded only once. The test images are ”Peppers,” ”Baboon,” ”Bar-
bara,” and ”Lena.” For fair comparison, the PSNR values of all the images are kept
consistent (about 42dB) for the two watermarking algorithms. We repeat our method
100 times with 100 different pseudo-random binary watermarks. The BER is calculated
by averaging the results of 100 times of the four test images. Fig. 3 shows two water-
marked images of the proposed watermarking algorithm. From the figure, we can see
that there is no visual distortion of the watermarked images.
Lossy JPEG compression is the most commonly image process in applications. The
results of robustness comparison against this attack are shown in Fig.4. It is clearly
demonstrated that the proposed improved RDM method outperforms RDM-Basic
method under JPEG compression attack, especially when the quality factor belows 30.
Fig. 5 shows the BER comparison under amplitude scaling attack. It can be seen that
both our proposed method and the RDM-Basic method are very robust to this attack
as expected. The nonlinear amplitude scaling, gamma correction, is also tested. The
results are shown in Table 1. It can be seen that the proposed improved RDM method
can significantly improve the performance against gamma correction.
(a) Original Barbara
(b) Original Lena
(c) Embedded Barbara
(d) Embedded Lena
Fig. 3. Test images
 
Search WWH ::




Custom Search