Int. J. Electron. Commun.

(AEÜ) 78 (2017) 64–71

Contents lists available at ScienceDirect

International Journal of Electronics and

Communications (AEÜ)
journal homepage: www.elsevier.com/locate/aeue

Review

A novel blind color image watermarking using upper Hessenberg matrix

Qingtang Su a,⇑, Beijing Chen b
a
School of Information Science and Engineering, Ludong University, 264025 Shandong, China
b
School of Computer & Software, Nanjing University of Information Science & Technology, 210044 Nanjing, China

a r t i c l e i n f o a b s t r a c t

Article history: It is a challenging work to design a blind color image watermarking scheme for protecting copyright,
Received 27 August 2015 which is different from the existing schemes used binary image or grayscale image as watermark and
Accepted 16 May 2017 is also different from other non-blind watermarking schemes. In this paper, we analyze the feature of
the upper Hessenberg matrix, and propose a blind color image watermarking scheme using upper
Hessenberg matrix of Hessenberg transform. Arnold transform is used to improve the security, and the
Keywords: MD5-based Hash pseudo-random algorithm is also used to improve the robustness. In the process of
Hessenberg transform
watermark embedding, the encrypted watermark information is embedded into the biggest energy ele-
Hessenberg matrix
Color image watermark
ment of the Hessenberg matrix by quantization technique. In the process of watermark extraction, the
Blind watermarking watermark is extracted from the attacked host image with blind manner. Simulation results show that
the proposed scheme outperforms other related methods in the aspects of the invisibility, robustness,
capacity and computational complexity.
Ó 2017 Elsevier GmbH. All rights reserved.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2. The Hessenberg transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3. Proposed method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.1. Embedding process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.2. Extraction watermark process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4. Experimental results and discussion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.1. The watermark invisibility analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.2. The watermark robustness analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.3. The capacity analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.4. The execution time comparison analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5. Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

1. Introduction [24–31]. Digital watermarking has been considered a potential

solution for providing further protection of digital content. The
With the widespread popularity of Internet and the rapid devel- close integration of the hidden signal, i.e., digital watermark, with
opment of multimedia technology, illegal copying, tampering, the host media (such as video, image, audio, and text) can be used
modifying digital copyright have become more and more impor- for declaring/verifying the ownership of the content, controlling
tant issues [1–3]. This phenomenon has led to an increasing need the software/hardware operations or for the trailer tracking pur-
for developing some standard solutions to prevent these issues pose [1].
Nowadays, some watermarking researches based on matrix
decomposition have been proposed [4–12]. For example, Lai [4]
⇑ Corresponding author.
designed a novel watermarking method based on human visual
E-mail address: sdytsqt@163.com (Q. Su).

http://dx.doi.org/10.1016/j.aeue.2017.05.025
1434-8411/Ó 2017 Elsevier GmbH. All rights reserved.
 Q. Su, B. Chen / Int. J. Electron. Commun. (AEÜ) 78 (2017) 64–71 65

system (HVS) and singular value decomposition (SVD), in which 2. The Hessenberg transform
the binary watermark was embedded into grayscale image with
size of 512
512 by modifying the certain coefficients of the uni- Hessenberg transform [8,19] is the factorization of a general
tary matrix U. This method has better performance in resisting the matrix A by orthogonal similarity transformations into the form
adding noise, cropping and median filtering, but is worse in the
aspect of resisting the rotation and scaling, and has false-positive A ¼ QHQ T ð1Þ
detection problem. Golea et al. [5] proposed a blind color image
where Q is an orthogonal matrix and H is an upper Hessenberg
watermarking by modifying the singular values, in which the
matrix, meaning that hij ¼ 0 whenever i > j þ 1. Hessenberg trans-
watermarked image was seriously affected because one or more
form is typically computed by Householder matrices. The House-
singular values would be modified to embed watermark. Bhatnagar
holder matrix (P) is an orthogonal matrix as follows.
et al. [6] embedded the grayscale watermark of size 256
256 into
the grayscale image of size 512
512. This method belongs to non- P ¼ ðIn  2uuT Þ=uT u ð2Þ
blind watermarking method and has false-positive detection prob-
n
lem. Naderahmadian et al. [7] proposed a grayscale image water- where u is a non-zero vector in R and In is the n
n identity matrix.
marking algorithm based on QR decomposition. It was found that There are n  2 steps in the overall procedure when A is of size
this method has lower computing complexity and better water- n
n. Therefore, Hessenberg transform is computed as
mark performance, but the embedded watermark was binary logo
H ¼ ðP1 P2    Pn3 Pn2 ÞT AðP 1 P2    P n3 P n2 Þ ð3Þ
of size 32
32. In [8], the grayscale image of size 64
64 was
embedded into grayscale image with size of 256
256 based on
Hessenberg transform, which belongs to non-blind method. Seddik ) H ¼ Q T AQ ð4Þ
et al. [9] proposed a blind watermarking method using Hessenberg
transform, in which the host image was grayscale image. Based on ) A ¼ QHQ T ð5Þ
QR decomposition, a watermark bit was embedded in all elements
where Q ¼ P1 P2    P n3 Pn2 .
of the first row of R matrix of size 8
8 by method [10], in which
For illustration, an image block A of size 4
4 and its Hessen-
the watermark was 88
88 binary image. In addition, a method
berg transform process is presented as follows.
[11] proposed to embed 32
32 binary image into 512
512 host
2 3
image by modifying the coefficients elements in the Q matrix of QR 80 91 91 95
decomposition. 6 83 89 88 96 7
6 7
At present, the trademarks or logos of many corporations are A¼6 7
4 90 89 89 96 5
colored and protecting copyright with color image has become a
more and more urgent problem. Although many color image 96 93 88 95
watermarking methods have been proposed [12–18], the embed- When A is decomposed by Hessenberg transform, its orthogonal
ded watermark is binary image or grayscale image when the color matrix Q and upper Hessenberg matrix H are given as follows,
image is used as the host image [12–16]. When the color image respectively.
watermark of the same size as binary image is embedded into color 2 3
host image, the information capacity of color image watermark 1 0 0 0
6 0 0:5335 0:7622 0:3667 7
will increase twenty-four times than that of binary image water- 6 7
Q ¼6 7
mark, which will directly affect the invisibility and robustness of 4 0 0:5785 0:0125 0:8156 5
watermark. In our previous works [1] and [18], two different 0 0:6170 0:6473 0:4476
watermarking schemes based on QR decomposition were pro-
posed, respectively. Although the latter method [18] has better 2 3
80:0000 159:8089 6:7321 1:6707
performance than that of the former method [1], the computa- 6 155:5796
tional complexities of these methods are higher. As well known, 6 273:8047 10:2233 6:7820 7
7
H¼6 7
the computational complexity of SVD or Schur decomposition is 4 0 15:1564 1:9211 0:2571 5
bigger than that of QR decomposition, and Hessenberg transform 0 0 1:6583 1:1164
is an intermediate step in QR decomposition. Therefore, Hessen-
In the above Hessenberg matrix H, the coefficient 273.8047 has
berg transform is of lower computational complexity than other
the biggest energy, which can be suitably modified to embed
decomposition methods and will be explored for watermarking
watermark information by quantization modulation.
technique.
Motivated by the above discussion, we propose a novel blind
color image watermarking scheme using upper Hessenberg matrix 3. Proposed method
obtained by Hessenberg transform to overcome those drawbacks.
By further analysis the 4
4 image block decomposed by Hessen- The proposed method mainly includes embedding watermark
berg transform, it is found that the biggest energy element of upper process and extraction watermark process.
Hessenberg matrix can be adopted to embed watermark informa-
tion. When extracting watermark, the original host image or the 3.1. Embedding process
original watermark image is not required. The simulation results
show that the proposed method has better performance in the The embedding watermark process using Hessenberg matrix is
aspects of invisibility, robustness, capacity and computational illustrated by Fig. 1, and its details are listed as follows.
complexity. S1: Convert the color image watermark to binary information.
The rest of this paper is organized as follow: an overview of the Firstly, the original color watermark image W of size p
q is
Hessenberg transform and the optimal embedding position in partitioned to three 2-D watermark layers by RGB color model,
Hessenberg matrix are presented in Section 2. The detailed embed- respectively. Secondly, each watermark layer is scrambled by
ding and extraction steps of the proposed method are introduced Arnold transform with private key KAm (m = 1, 2, 3) for improving
in Section 3. In section 4, the simulation results and discussion the security of the watermarking [20]. Thirdly, each pixel value is
are shown. Section 5 gives the conclusions of this paper. further converted to 8 bits binary sequence. Finally, all binary
66 Q. Su, B. Chen / Int. J. Electron. Commun. (AEÜ) 78 (2017) 64–71
Fig. 1. Diagram of the watermark embedding process.
sequences are combined to obtain the binary watermark layer Wj
(j = 1, 2, 3).
S2: Preprocess the host image.
For embedding watermark, the host image I is also partitioned
to three image layers of R, G and B, i.e. Ij (j = 1, 2, 3), and each image
layer Ij is further divided into non-overlapping image blocks with
size 4
4.
S3: Select the random blocks for embedding watermark.
For improving the robustness of anti-cropping of the proposed
method, the blocks for embedding watermark is dispersedly
selected instead of centralized embedding. The Hash pseudo-
random replacement algorithm, which based on MD5 with private
key KBn (n = 1, 2, 3), is used to randomly select the embedding
image blocks from image layer Ij (j = 1, 2, 3) [21], and only half of
all blocks are selected in this proposed method.
S4: Perform Hessenberg transform.
According to Eq. (1), each selected image block is executed by
Hessenberg transform and its upper Hessenberg matrix H is gotten.
S5: Embed watermark.
Search for the biggest energy element hmax of the upper Hessen-
berg matrix H, and the watermark w is embedded into the selected

block by modifying hmax with hmax according to the following Eq. (6).

 hmax  modðhmax ; TÞ þ 0:75 T; if w ¼ ‘1’
hmax ¼
hmax  modðhmax ; TÞ þ 0:25 T; if w ¼ ‘0’
where modð:Þ is the module operation, and T is the quantization
step based on private key KT.
S6: Perform inverse Hessenberg transform.
Fig. 2. Diagram of the watermark extracting process.
Use Eq. (7) to inverse Hessenberg transform and obtain the
watermarked image block A .
A ¼ QH Q T ð7Þ
S7: Loop
The above-mentioned steps S4-S6 are repeated to embed all
watermark bits into three image layers Ij (j = 1, 2, 3). At last, the
watermarked image I⁄ is recombined by the watermarked image
layers R, G, B.
3.2. Extraction watermark process
The watermark extracting process of the proposed method is
illustrated by Fig. 2. As can be seen from it, the original host image
or the watermark image is not required when extracting water-
mark. So, this proposed method belongs to blind watermarking
scheme. The extraction watermark process is introduced as follows.
S1: Preprocess the watermarked image.
At first, the watermarked image I⁄ is partitioned to the three
image layers of R, G, B. Then, each image layer is further divided
into non-overlapping image blocks of size 4
4.
S2: Select the watermarked image block.
The MD5-based Hash pseudo-random replacement algorithm
ð6Þ
with private key KBn (n = 1, 2, 3) is used to select the watermarked
image blocks.
S3: Performe Hessenberg transform.
Each selected image block is decomposed by Hessenberg trans-
form and its upper Hessenberg matrix H⁄ is gotten.
 Q. Su, B. Chen / Int. J. Electron. Commun. (AEÜ) 78 (2017) 64–71 67

S4: Extract watermark. 2552

 PSNR ¼ 10  lg ð9Þ
Search for the biggest energy element hmax of the upper Hessen- MSE
berg matrix H⁄, then use the following Eq. (8) to extract the water-
mark information W⁄ with the quantization step T that based on where MSE is the mean square error given by
private key KT.
1 X 3 X m X n
  MSE ¼ ½Iðx; y; jÞ  I ðx; y; jÞ
2
ð10Þ
‘0’; if modðhmax ; TÞ < 0:5  T 3MN j¼1 x¼1 y¼1
W ¼ ð8Þ
‘1’; else
where Iðx; y; jÞ, I ðx; y; jÞ present the value of pixel (x, y) in layer j of
The above-mentioned steps S2-S4 are repeated to extract
the original image and the watermarked one, and m, n denote the
watermark information from all watermarked image blocks.
width and the height of the host images, respectively.
S5: Obtain the watermark layer.
Moreover, SSIM proposed in [23], which has correlation with
The extracted information W⁄ is divided into some binary
the quality perception of the human visual system (HVS), is defined
sequences of length 8 bits, and each binary sequence is converted
as
to decimal number, then the watermark layer W j (j = 1, 2, 3) is
formed. SSIMðI; I Þ ¼ lðI; I ÞcðI; I ÞsðI; I Þ ð11Þ
S6: Obtain the final watermark.
 
At first, each watermark layer W j (j = 1, 2, 3) is transformed by where lðI; I Þ is luminance comparison function, cðI; I Þ is contrast
the anti-Arnold transformation with the private key KAm (m = 1, 2, comparison function, and sðI; I Þ is structure comparison function.
3). Then, the final extracted watermark W⁄ is combined with three Besides, in order to measure the robustness of the watermark,
extracted watermark layers. we use the normalized correlation (NC) between the original
watermark W and the extracted watermark W  , which is denoted
as follows
4. Experimental results and discussion
P3 Pp Pq 
j¼1 x¼1 y¼1 ðWðx; y; jÞ  W ðx; y; jÞÞ
In this paper, the performance of the proposed method is mea- NC ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P3 Pp Pq ffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P3 Pp Pq
2  2
sured by the invisibility, robustness, capacity, and computational j¼1 x¼1 y¼1 Wðx; y; jÞ j¼1 x¼1 y¼1 W ðx; y; jÞ
complexity. To validate the watermark performances of the pro- ð12Þ
posed method, all 24-bit 512
512 color images of the image data-
base CVG-UGR are used as the host images [22], and two 24-bit where Wðx; y; jÞ, W  ðx; y; jÞ present the value of pixel (x, y) in layer j
32
32 color images shown in Fig. 3 are used as original water- of the original watermark and the extracted one, and P, Q denote the
marks. For relatively fair comparison, the host images in Fig. 4 used row and the column size of the original watermark image,
in [5,18] are adopted in this experiment. All the schemes are coded respectively.
in MATLAB 2010a and performed on a laptop computer with Intel
Core CPU @ 2.27GHZ, 2.00 GB RAM.
For measuring the invisibility, two metrics that includes peak
signal-to-noise ratio (PSNR) and structural similarity index mea-
surement (SSIM) are adopted to measure the objective similarity
between the original color host image I and the watermarked color
image I :
As a good rule for the watermark visibility estimation, PSNR is
described by

Fig. 5. The results of invisibility and robustness with different quantization steps.

Table 1
The average PSNR, SSIM and NC values after using different watermark images under
no-attacks.

Watermark PSNR(dB) SSIM NC

(a) (b)
Watermark Fig. 3(a) 37.3922 0.9426 0.9984
Watermark Fig. 3(b) 37.5258 0.9438 0.9979
Fig. 3. Watermark image: (a) PEUGEOT logo, (b) 8-color image.

Fig. 4. (a-d) Test image: Lena, F16, Peppers, and House, respectively.
68 Q. Su, B. Chen / Int. J. Electron. Commun. (AEÜ) 78 (2017) 64–71

Table 2
The PSNR(dB) and SSIM for various watermarking methods.

Image Scheme [5] Scheme [10] Scheme [11] Scheme [18] Proposed Scheme
Lena 39.4358/0.9935 41.3529/0.9767 22.5616/0.6332 35.3521/0.9589 36.3947/0.9371
F16 38.3922/0.9540 41.4073/0.9755 20.4106/0.5411 36.3160/0.9256 37.4429/0.9321
Peppers 34.4587/0.9279 41.3701/0.9631 23.2864/0.7111 36.6869/0.9682 37.3639/0.9342
House 34.4806/0.9970 34.4642/0.9154 25.6319/0.9249 34.4806/0.9970 37.6319/0.9249

Fig. 6. The extracted watermarks from the watermarked image ‘‘Lena” by different methods under different attacks.
 Q. Su, B. Chen / Int. J. Electron. Commun. (AEÜ) 78 (2017) 64–71 69

Fig. 7. The extracted watermarks from the watermarked image ‘‘F16” by different methods under different attacks.

In the proposed method, the quantization step plays an impor-
tant role. For deciding the quantization step T, standard images
selected from image database CVG-UGR are repeatedly simulated.
As can be seen from Fig. 5, with the increase of the quantization
step, the average SSIM value is decreasing but the average NC value
is increasing, which means the watermark invisibility is weaker
and weaker but the watermark robustness gets stronger. Consider-
ing the tradeoff between the invisibility and the robustness of
watermark, the range of the quantization step T is set between
63 and 68. In this paper, the quantization step is set to 65 for test-
ing the related watermarking performance.
4.1. The watermark invisibility analysis
In the proposed method, two 24-bit color image watermarks are
embedded into the 24-bit color host images from image database
CVG-UGR, respectively. Table 1 lists the average PSNR and SSIM
values of all watermarked images. The experimental data shows
that the watermarking method has better watermark invisibility.
Meanwhile, the average NC values of extracted watermark images
are also given to prove the effectiveness of extraction process.
For further showing the invisibility, we compare the proposed
method with other related methods of [5,10,11,and18]. Table 2
70 Q. Su, B. Chen / Int. J. Electron. Commun. (AEÜ) 78 (2017) 64–71

gives comparison results of different watermarking schemes in Table 3

terms of PSNR and SSIM. From Table 2, we can know that the invis- The comparison of watermark embedding capacity between different methods.

ibility of scheme [11] based on QR decomposition is not good. Method Watermark Length (Bit) Host image (Pixel) Bit/Pixel
Hence, it cannot satisfy the needs of color watermarking. Rela- Scheme [5] 32
32
24 512
512
3 0.03125
tively, these schemes including scheme [5] based on SVD, scheme Scheme [10] 88
88 512
512 0.02954
[10] based on QR decomposition, scheme [18] based on QR decom- Scheme [18] 32
32
24 512
512
3 0.03125
position and the proposed one have better invisibility because Proposed 32
32
24 512
512
3 0.03125

their PSNR values are all more than 30 dB. For a 4
4 matrix, there
are at least 8 elements will be modified using the above methods to
embed a watermark. But, as can be seen from Eq. (5), there are 6
elements in matrix A will be modified when one element of H
matrix is modified, which is one of the main reasons for further
improving the watermark invisibility of this proposed method.
4.2. The watermark robustness analysis
In digital image processing, the watermarked image is often
attacked by compressing image, noising image, scaling image,
filtering image, cropping image, blurring image etc. In this experi-
ment, the watermarked image is attacked by lossy JPEG and JPEG
2000 compression range from 0 to 100, Gaussian noise with mean
0 and variance from 0.001 to 0.01, salt & peppers noise with noise
density from 0.02 to 0.1, median filter with size from 2
2 to 5
5,
low-pass filter with order 100 and cut-off frequency from 1db to
5db, sharpen with radius from 0.2 to 1, burring with radius from
0.2 to 1, scaling ratio from 0.25 to 4 (in which the watermarked
image of size 512
512 is scaled up or down by the scaling ratio,
and then resize the size of scaled image to 512
512), and
cropping range from 20% to 50%. For saving some space, Figs. 6
and 7 show the extracted image watermark from the attacked
images ‘‘Lena” and ‘‘F16” after embedding the watermark of
Fig. 3(a), respectively. As can be seen from Figs. 6 and 7, the pro-
posed method has obvious advantage than other methods when
the watermarked image is attacked by JPEG with compression fac-
tor 30. In addition, the visual effect of extracted watermark by the
proposed method is better than other methods when perform
cropping attack, which because the MD5-based Hash pseudo-
random replacement algorithm with private key is used to select
randomly the embedding position of watermark. In sum, the pro-
posed scheme is better than other schemes [5,10,18] under most
attacks in the aspect of the robustness.
4.3. The capacity analysis
In this paper, the watermark with length of (512
512
3)/
(4
4) = 49,152 (bits) can be embedded into the 24-bit color host
image with size of 512
512, which is the theory limit. But, the
quality of watermarked image will be seriously degraded when
reaching the theory limit of watermark. So, this paper uses 24-bit
color image of size 32
32 as watermark which length is 24,576
bits (i.e., 32
32
24). Table 3 shows the comparison results of
watermark capacity. In methods [5,18] and the proposed method,
the embedded block sizes are all 4
4, the capacity is
ð32
32
24Þ=ð512
512
3Þ ¼ 0:03125 ðBit=PixelÞ ð13Þ
Table 4
The comparison of execution time between different methods (Second).
Method Embedding time Extraction time Total time
Scheme [5] 1.909066 0.905951 2.815017
Scheme [10] 0.913247 0.455445 1.368692
Scheme [18] 0.829863 0.651540 1.481403
Proposed 0.566331 0.320511 0.886842
tion time of the proposed method is less than that of scheme [5]
based on SVD and scheme [18] based on QR decomposition. The
proposed costs less time because it uses only Hessenberg trans-
form, while the scheme [10] includes Wavelet transform and QR
decomposition.
5. Conclusions
A novel blind color image watermarking algorithm using upper
Hessenberg matrix is proposed in this paper. Based on the Hessen-
berg transform, the color image watermark information is embed-
ded into the maximum element in the upper Hessenberg matrix by
modifying the biggest energy element. Moreover, the embedded
watermark information can be blindly extracted from the attacked
images. Simulation results show that the proposed scheme outper-
forms other related methods in the aspects of the invisibility,
robustness, capacity and computational complexity. However,
the performance of resisting image rotating attack will be further
considered in future work.
Acknowledgements
The research was partially supported by the Natural Science
Foundation of China (61472172, 61572258, 61602229), Priority
Academic Program Development of Jiangsu Higher Education Insti-
tutions (PAPD), Jiangsu Collaborative Innovation Center on Atmo-
spheric Environment and Equipment Technology (CICAEET),
Natural Science Foundation of Shandong Province (ZR2016FM21,
ZR2016FM13), Doctoral Foundation of Ludong University
(LY2014034), Shandong Province Science and Technology Plan Pro-
jects (2013YD01031, 2015GSF116001), Key Science and Technol-
ogy Plan Projects of Yantai City (2016ZH057). The authors would
like to thank anonymous referees for their valuable comments
and suggestions which lead to substantial improvements of this
paper.

Since the block size in [10] is 8
8, the capacity in methods

[5,18] and the proposed method is higher than 0.02954 Bit/Pixel
of method [10].
4.4. The execution time comparison analysis
It is well known from Table 4 that the execution time of the pro-
posed algorithm is less than that of others. Theoretically, the com-
putational complexity of SVD or Schur decomposition is more than
QR decomposition, and Hessenberg transform is also an intermedi-
ate step of QR decomposition. Hence, the embedding and extrac-
References
[1] Su Q, Niu Y, Wang G, Jia S, Yue J. Color image blind watermarking scheme
based on QR decomposition. Signal Process 2014;94(1):219–35.
[2] Li J, Li X, Yang B, Sun X. Segmentation-based image copy-move forgery
detection scheme. IEEE Trans Inf Forensics Secur 2015;10(3):507–18.
[3] Zheng Y, Jeon B, Xu D, Wu QMJ, Zhang H. Image segmentation by generalized
hierarchical fuzzy C-means algorithm. J Intell Fuzzy Syst 2015;28(2):961–73.
[4] Lai CC. An improved SVD-based watermarking scheme using human visual
characteristics. Opt Commun 2011;284(4):938–44.
[5] Golea NEH, Seghir R, Benzid R. A bind RGB color image watermarking based on
singular value decomposition. In: 2010 IEEE/ACS international conference on
computer systems and applications (AICCSA). p. 1–5.
 Q. Su, B. Chen / Int. J. Electron. Commun. (AEÜ) 78 (2017) 64–71
[6] Bhatnagar G, Raman B. A new robust reference logo watermarking scheme.
Multimedia Tools App 2011;52(2–3):621–40.
[7] Naderahmadian Y, Hosseini-Khayat S. Fast and robust watermarking in still
images based on QR decomposition. Multimedia Tools App 2013;72
(3):2597–618.
[8] Bhatnagar G, Wu QMJ. Biometircs inspired watermarking based on a fractional
dual tree complex wavelet transform. Future Gener Comput Syst 2013;29
(1):182–95.
[9] Seddik H, Sayadi M, Fnaiech F, Cheriet M. Image watermarking based on the
Hessenberg transform. Int J Image Graphics 2009;9(03):411–33.
[10] Yashar N, Saied HK. Fast watermarking based on QR decomposition in Wavelet
domain. In: 2010 Sixth international conference on intelligent information
hiding and multimedia signal processing. p. 127–30.
[11] Song W, Hou J, Li Z, Huang L. Chaotic system and QR factorization based robust
digital image watermarking algorithm. J Cent South Univ Technol 2011;18
(1):116–24.
[12] FindIk O, Babaoglu I, Ülker E. A color image watermarking scheme based on
artificial immune recognition system. Expert Syst Appl 2011;38(3):1942–6.
[13] Niu PP, Wang XY, Yang YP, Lu MY. A novel color image watermarking scheme
in nonsampled contourlet-domain. Expert Syst Appl 2011;38(3):2081–98.
[14] Vahedi E, Zoroofi RA, Shiva M. Toward a new wavelet-based watermarking
approach for color images using bio-inspired optimization principles. Digital
Signal Process 2012;22(1):153–62.
[15] Wang X, Wang C, Yang H, Niu P. A robust blind color image watermarking in
quaternion Fourier transform domain. J Syst Softw 2013;86(2):255–77.
[16] Shao Z, Duan Y, Coatrieux G, Wu J, Meng J, Shu H. Combining double random
phase encoding for color image watermarking in quaternion gyrator domain.
Opt Commun 2015;343:56–65.
[17] Chen B, Coatrieux G, Chen G, Sun X, Coatrieux JL, Shu H. Full 4-D quaternion
discrete Fourier transform based watermarking for color images. Digital Signal
Process 2014;28(5):106–19.
[18] Su Q, Niu Y, Zou H, Zhao Y, Yao T. A blind double color image watermarking
algorithm based on QR decomposition. Multimedia Tools App 2014;72
(1):987–1009.
71
[19] Golub GH, Loan CFV. Matrix computations. Baltimore: Johns Hopkins
University Press; 1996.
[20] Chen W, Quan C, Tay CJ. Optical color image encryption based on Arnold
transform and interference method. Opt Commun 2009;282(18):3680–5.
[21] Rivest RL. The MD5 message digest algorithm, Internet RFC 1321; April 1992.
[22] University of Granada, Computer Vision Group. CVG-UGR Image Database.
[2012-10-22]. http://decsai.ugr.es/cvg/dbimagenes/c512.php. 20.
[23] Wang Z, Bovik AC, Sheikh HR, Simoncelli EP. Image quality assessment: from
error visibility to structural similarity. IEEE Trans Image Process 2004;13
(4):600–12.
[24] Ma T, Zhou J, Tang M, Tian Y, Abdullah AD, Mznah AR, Sungyoung L. Social
network and tag sources based augmenting collaborative recommender
system. IEICE Trans Inf Syst 2015;98(4):902–10.
[25] Xia Z, Wang X, Sun X, Wang B. Steganalysis of least significant bit matching
using multi-order differences. Secur Commun Netw 2014;7(8):1283–91.
[26] Zhou Z, Wang Y, Wu QMJ, Yang C, Sun X. Effective and efficient global context
verification for image copy detection. IEEE Trans Inf Forensics Secur 2017;12
(1):48–63.
[27] Xia Z, Wang X, Zhang L, Qin Z, Sun X, Ren K. A privacy-preserving and copy-
deterrence content-based image retrieval scheme in cloud computing. IEEE
Trans Inf Forensics Secur 2016;11(11):2594–608.
[28] Fu Z, Wu X, Guan C, Sun X, Ren K. Toward efficient multi-keyword fuzzy search
over encrypted outsourced data with accuracy improvement. IEEE Trans Inf
Forensics Secur 2016;11(12):2706–16.
[29] Fu Z, Ren K, Shu J, Sun X, Huang F. Enabling personalized search over encrypted
outsourced data with efficiency improvement. IEEE Trans Parallel Distrib Syst
2016;27(9):2546–59.
[30] Zhou Z, Yang C, Chen B, Sun X, Liu Q, Wu QMJ. Effective and efficient image
copy detection with resistance to arbitrary rotation. IEICE Trans Inf Syst 2016;
E99-D(6):1531–40.
[31] Liu Q, Cai W, Shen J, Fu Z, Liu X, Linge N. A speculative approach to spatial-
temporal efficiency with multi-objective optimization in a heterogeneous
cloud environment. Secur Commun Netw 2016;9(17):4002–12.