行政院國家科學委員會專題研究計畫 成果報告
植基於最佳化演算法之多用途數位浮水印技術
計畫類別： 個別型計畫 計畫編號： NSC93-2213-E-151-016- 執行期間： 93 年 08 月 01 日至 94 年 07 月 31 日 執行單位： 國立高雄應用科技大學電子工程系 計畫主持人： 潘正祥 共同主持人： 謝欽旭 報告類型： 精簡報告 報告附件： 出席國際會議研究心得報告及發表論文 處理方式： 本計畫涉及專利或其他智慧財產權，2 年後可公開查詢中 華 民 國 94 年 10 月 26 日
行政院國家科學委員會專題研究計畫成果報告
植基於最佳化演算法之多用途數位浮水印技術
Multipurpose Digital Watermarking Technique Based on Optimal
Algorithms
計畫編號：NSC 93-2213-E-151 -016
執行期限：93 年 8 月 1 日至 94 年 7 月 31 日
主持人：潘正祥
國立高雄應用科技大學電子工程系
E-mail : jspan@cc.kuas.edu.tw
一、 中文摘要 由於數位科技的發展，複製的程序越 來越簡單，因此數位影像很容易被複製， 甚至被竄改。有鑑於此，我們可利用數位 浮水印技術來解決此問題，證明數位影像 的版權所有及保障智慧財產權。在本專案 計劃中，我們提出一多重浮水印技術針對 多媒體資料進行保護。我們首先分析並介 紹過去許多植基於 Spatial 及 DCT/Wavelet 視覺模組方法的優缺點。此外，為了克服 先前方法的缺點並自適應於各種影像攻 擊，我們嵌入兩個扮演功能互補角色的浮 水印，以期在遭遇任何攻擊情況下，至少 還有一個浮水印能存活。以可見的浮水印 宣告版權擁有人，並且用不可見的浮水印 作為版權保護。 這兩個浮水印運用不同 的技術嵌入在影像中。此外提出一個最佳 化技術，以禁止搜尋演算法實現浮水印的 強健性與改善影像品質。實驗結果證明我 們所提的方法能有效改善數位浮水印技 術中，影像品質與強健性浮水印相互制衡 的問題。 關鍵詞：數位浮水印、數位資料保護、版 權保護、版權宣告、強健浮水印 AbstractIn this project, we presented an image watermarking method for two purposes: one is to notify the copyright owner with a visible watermark, and the other one is to protect the copyright with an invisible watermark. These two watermarks are embedded in different domains by using different techniques. An innovative watermarking scheme based on tabu search (TS) algorithms in both the spatial domain and the transform domain is proposed. The fidelity performance of an existing image watermarking scheme is enhanced through the TS optimization. The experimental results show that the proposed techniques can improve the degraded image quality greatly. Simulation results show that the proposed method may improve the PSNR for embedding the visible watermark and enhance the robustness of the invisible watermark.
Keywords : Digital Watermark 、 Digital
Data Protection 、 Copyright Protection 、 Copyright Declaration 、 Robust Watermark
二、 緣由與目的
With the widespread use of Internet and the development in computer industry, the digital media, including images, audios, and video, suffer from infringing upon the copyrights with the digital nature of unlimited copying, easy modification and quick transfer over the Internet. In this project, we concentrate our research topic on image watermarking for copyright protection.
Digital watermarking for images is one way to embed the secret information, or the watermark, into the original image itself to protect the ownership of the original sources [1]. In general, they can be classified into visible and invisible watermarks [2-3]. A visible watermark can be used to provide instantaneous recognition of the ownership, prevent the unauthorized use of publicly available images commercially [4-8]. The invisible watermark can be used for copyright protection or content authentication. The visible watermarks are easily identified. They are usually not robust against image cropping. The invisible watermarks are more secure and robust than the visible watermarks. The embedding locations are secret, and only the authorized persons with the secret keys in the watermarking system can extract the secret watermark.
A visible watermark and an invisible watermark can be embedded in different parts of the discrete cosine transform (DCT) and discrete wavelet transform (DWT) coefficients of the host images. In our
design, the visible watermark is embedded in the spatial domain, whereas the invisible watermark is embedded in the wavelet domain. The tabu search approach [9] is applied to select the optimal spatial position to embed the visible watermark based on the improvement of the PSNR and the suitable bands of the wavelet transform to embed the invisible watermark so as to improve security, robustness, and image quality of the watermarking system.
Here we present the watermarking algorithm to illustrate the method we presented in the project. For convenience, let the original image X and the visible watermark V be 256 gray-level image of size 256x256, and 64x64, respectively, and the invisible watermark W be a binary image of size32x32. The embedding algorithm can be described as follows:
A. Visible Watermark Embedding
1. Calculate the variance vkl for each possible image block of size 8x8. Find the maximal variance vmax, and the minimal variance vmin, and calculate the normalized variance by min max min v v v vkl kl − − = ϕ (1)
2. Perform the visible watermarking process in the spatial domain based on the following equation, which can be
described as:
(
)
ij ij kl ij kl ij V X X X 255 1 ' =ϕ ⋅ + −ϕ (2)where Xij' , X and ij V denote the ij pixels at position (ij) in the block (k,l) of the visibly watermarked image, the original image and the visible
watermark, respectively. ϕij is the normalized variance of pixels in the original image block (k,l).
3. Apply the tabu search approach to find the embedding position in the spatial domain such that the PSNR between the original image and the watermarked image with the visible watermark can be maximized.
B. Invisible Watermark Embedding
The basic idea of the embedding algorithm for invisible watermark is similar to [12], except it is optimized by tabu search approach and applied in the wavelet domain.
1. Discrete wavelet transform (DWT) is an efficient tool for multi-resolution analysis of signal [10], especially in image processing [11]. Among many DWTs, the Haar discrete wavelet transform (Haar DWT) is the simplest and orthogonal one. Before the embedding procedure, we need to transform the spatial domain pixels into the frequency bands of the Haar DWT domain. We perform the 8 x 8 block DWT on the image X' with size M×N first and get the coefficients in the frequency bands, Y, Y =DWT
( )
X' (3) andUU
(
)
8 1 8 1 , M m N n n m Y Y = = = (4) For one non-overlapping block (m, n) in X', the resulting 64 DWT bands Y(m, n) can be represented by(
)
U
63{
(
)( )
}
0 , , = = k k n m Y n m Y , 8 1≤m≤ M , 8 1≤n≤ N (5) Afterwards, we are able to embed thewatermark in the DWT domain.
2. Assuming that the binary-valued watermark to be embedded is W with size Mw×Nw. The spatial positions of the
watermark bits are permuted using pseudo-random number traversing method. With a pre-determined key, key0 in the pseudo-random number generation system, we have the
permuted watermark WP,
(
W, key0)
permuteWP = (6) The watermark bits WP, are embedding into the selected DWT frequency bands. 3. Assume the set of the frequency band be
F. By applying the tabu search approach, the DWT frequency bands are chosen in F for WP to match both the watermarked image quality and the robustness under certain attacks in the training iteration. The invisible watermarking process in the frequency domain based on the following equation : ( )
( )
( )( ) ( )( )
( )( )
i if W( )( ) i F Y i Y F i W if i Y i n m p n m n m i n m p n m ∈ = = ∈ = + , 0 , 1 , , , ' , , θ (7)θ was set to be 3 in our experiment. 4. Perform the inverse DWT on Y' as
( )
' ''_DWT Y inverse
X = (8) 5. The mean squared error (MSE) between
the watermarked image with visible watermark and watermarked image with both visible and invisible watermarks can be defined by
( )
( )
(
)
∑∑
= = − ⋅ = M i N j j i X j i X N M MSE 1 1 2 '' ' , , 1 (9)Consequently, the watermarked image quality represented by the peak signal-to-noise ratio (PSYR) between X'
and X'' can be expressed by ⎟⎟ ⎠ ⎞ ⎜⎜ ⎝ ⎛ ⋅ = MSE PSNR 2 10 255 log 10 dB (10)
6. Assume X''p is the result of the JPEG attack for watermarked imageX''. For the preliminary experiment, JPEG attack is employed. We extract the watermark from X''p, and calculate the bit-correlation ratio (BCR) value between the embedded watermark and the extracted one. The BCR between the embedded watermark Wi,j, and the extracted watermark Wi,j' is defined by
(
)
w w M i N j ij ij N M W W BCR w w × ⊕ =∑∑
=1 =1 ' (11)7. After obtaining the PSNR in the watermarked image and the BCR value after IPEG attack, we are ready to start the TS training process. According to the definition of TS, we need to assign the fitness function in the each iteration with fc =PSNR+λ×BCRJPEG (12) where fc, and λ are the fitness value and the weighting factor for the BCR value. For simplicity, we use λ=10. C. The Extraction Algorithm
1. We are able to extract the permuted watermark, ( )
( )
( )( )
( )( )
otherwise i W i Y i Y if n m p n m n m 0 2 1 , , ' , = ≥ − θ (13) ( )( )
U U
MW W M m N N n n m p p W i i F W 1 1 ' , ' = = ∈ = (14)2. Finally, we use key0 in Eq. (6) to acquire the extracted watermark W’ from W’p,
(
0)
' ' , _ permuteW key inverse W = p (15) 三、 結果與討論In our simulation, we take the well-known test image, Lena, with size 256 × 256, as the original source, which is shown in Fig. 1. The visible watermark with size 64×64 and the invisible watermark with size 32×32 are shown in Fig. 2 and Fig. 3.
Fig. 1. The original test image Lena X
Fig. 2. Visible watermark V
Fig. 3. Invisible watermark W
The comparison of the random method and the TS training is shown in table 1. The watermarked image which is embedded the dual watermarks is shown in Fig. 4. The cropping experimental result is shown in Fig. 5. The PEG compression with quality factor 30% - 90% attacks are simulated and shown in Fig. 6 (a)-(g).
Fig. 4. X’’ embeds both the visible and invisible watermarks (PSNR=33.3245028692)
Fig. 5. PSNR=13.8537 BCR=0.96972
Watermarked image attached by cropping (64×
64)
Fig. 6. Invisible watermarks extracted from the attacked watermarked images
Table 1. The PSNR for using the TS and the
random number generation
四、 計畫成果自評
The purpose of this project aims at presenting an innovative watermarking scheme which not only notifies the copyright but also protects it. Under this project, we present a watermarking scheme combines the visible and invisible watermark in both special and frequency domain with the algorithm, which was presented previously. The results of the algorithm presented conform to the idea of this project. The algorithm, which presented in this project, has also presented in 2004 IEEE Asia-Pacific Conference on Circuits and Systems. Due to the support from this project, 17 conference papers and 5 journal papers are published.
五、 參考文獻
[1] S. Katzenbeisser and F. Petitcolas, Information Hiding – Techniques for Steganography and Digital Watermarking, Artech House, Norwood, 2000.
[2] C. S. Lu, and H. M. Liao, “Multipurpose Watermarking for image authentication and protection,” IEEE Trans. On ImageProcessing, vol. IO, no. IO, pp. 1579-1592,2001.
[3] Z. M. Lu, H. T. Wu, D. G. Xu and S. H: Sun, “A multipurpose image watermarking method for copyright notification and protection,” IEICE Trans. Inf and Syst., vol. E86-D, no. 9,pp. 1931-1933,2003.
visible video watermarks in the compressed domain,” Proc. of IEEE Int. Conference on Image Processing (ICIP’98). pp. 474-477, 1998.
[5] M. S. Kankanhalli, Rajmohan, and K. R. Ramakrishnan, “Adaptive visible watermarking of images,” Proc. oJIEEE Int. Conference on Mulfimedia Computing and Systems, pp. 568-573, 1999.
[6] P. M. Chen, “A visible watermarking mechanism using a statistic approach,” Proc. oJIEEE 5th Inf. Conf on Signal Processing (WCCC-ICSPZOOO), pp. 910-913,2000.
[7] S. P. Mohanty, K. R. Ramakrishnan, and M. S. Kankanhalli, “ A DCT domain visible watermarking technique for images,” Proc. of IEEE Int. Conference on Multimedia and
EXpo(ICUE’2000), pp. 1029-1032,2000.
[8] Y. J. Hu, and S. KWONG, “Wavelet
domain adaptive visible watermarking,” Electron. Lett., vol. 37,
no. 20, pp. 1219-1220,2001.
[9] F. Glover and hi. Laguna, Tabu search. Boston, MA: Kluwer Academic Publishers, 1997.
[10] S. G. Mallat, “A Theory for multiresolution signal decomposition: the wavelet representation,” IEEE Trans. On PAMI, vol. I I , no. 7, pp. 674-693, 1989.
[11] M. Antonini, M. Barlaud, P. Mathieu and I Daubechies, "Image Coding Using Wavelet Transform, " IEEE Trans on Image Processing, vol. I , no. 2, pp. 205-220, 1992.
[12] C. S. Shieh, H. C. Huang, F. H. Wang, and J. S. Pan, "Genetic Watermarking Based on Transform Domain Techniques," Pattern Recognition, vol. 37, no. 3, pp. 555465,2004.
