Improved RobustWatermarking Based on Rational Dither Modulation - Advances in Image and Graphics Technologies - page 312

Ta b l e 3 . BER under noise addition

Noise addition Proposed RDM-Basic10 RDM-Basic30 RDM-Basic50

Gaussian noise ( ˃ =10) 0.0323 0.1249 0.0942 0.0804

Gaussian noise ( ˃ =20) 0.2401 0.4344 0.4405 0.4400

Salt&Pepper ( p =0 . 01) 0.0512 0.0642 0.0662 0.0579

Salt&Pepper ( p =0 . 02) 0.0562 0.0660 0.0699 0.0608

Ta b l e 4 . BER under geometric attacks

Attacks Proposed RDM-Basic10 RDM-Basic30 RDM-Basic50

Rotation (0 . 5 ◦ )

0.3142 0.3367 0.3667 0.3629

− 0 . 5 ◦ )

Rotation ( 0.3123 0.3511 0.3801 0.3763

Resizing (256 × 256) 0.0037 0.0079 0.0056 0.0050

Resizing (128 × 128) 0.0603 0.0814 0.0699 0.0676

Ta b l e 5 . BER comparison between the proposed method and MWT-EMD [13]

Attacks Proposed MWT-EMD Attacks Proposed MWT-EMD

Median filtering (5 × 5) 0.0046 0.0975 JPEG (Q=10) 0 0

Median filtering (7 × 7) 0.0353 0.1094 JPEG (Q=20) 0 0

Rotation (1 . 0 ◦ )

Median filtering (9 × 9) 0.0896 0.6524 0.3250 0.5469

Rotation (0 . 5 ◦ ) Average filtering

(3 × 3)

0 - 0.1009 0.4492

Average filtering
(5 × 5)

0 - Rotation

( − 0 . 5 ◦ )

0.1110 0.4414

Gaussian filtering
(3 × 3)

0 0 Rotation

( − 1 . 0 ◦ )

0.2873 0.5703

Gaussian filtering
(5 × 5)

0 0.0156 Salt&Pepper

( p =0 . 08)

0 0.0284

4.2 Comparison with Other Watermarking Methods

In order to further evaluate the performance of the improved RDM method, we also
compare it with two watermarking methods MWT-EMD [13] and GDWM [10]. MWT-
EMD is the state-of-the-art method in spread spectrum watermarking and GDWM is
one of the state-of-the-art methods in quantization-based watermarking.
Table 5 compares the BER results of the improved RDM method with MWT-EMD
method. As in [13], the test images are ”Baboon,” ”Goldhill,” ”Lena,” and ”Pepper”
and a 64-bit message is embedded in each image with the PSNR of about 42dB. The
results of our method are the averaged BERs obtained from embedding 100 different
watermarks in each image. It can be seen that our method outperforms MWT-EMD
under all the considered attacks.
Table 6 compares the BER results of the improved RDM method with GDWM
method. As in [10], the test images are ”Baboon,” ”Barbara,” ”Lena,” and ”Pepper” and
a 256-bit message is embedded in each image with the PSNR of 43.29dB, 42.70dB,
43.54dB and 43.06dB respectively. We can see that our method is more robust than
GDWM in general.