On the implementation of a copyright

protection scheme using digital image

watermarking

Subhrajit Sinha Roy, Abhishek Basu &

Avik Chattopadhyay

Multimedia Tools and Applications

An International Journal

ISSN 1380-7501
Volume 79
Combined 19-20

Multimed Tools Appl (2020)

79:13125-13138
DOI 10.1007/s11042-020-08652-9

1 23
Your article is protected by copyright and
all rights are held exclusively by Springer
Science+Business Media, LLC, part of
Springer Nature. This e-offprint is for personal
use only and shall not be self-archived in
electronic repositories. If you wish to self-
archive your article, please use the accepted
manuscript version for posting on your own
website. You may further deposit the accepted
manuscript version in any repository,
provided it is only made publicly available 12
months after official publication or later and
provided acknowledgement is given to the
original source of publication and a link is
inserted to the published article on Springer's
website. The link must be accompanied by
the following text: "The final publication is
available at link.springer.com”.

1 23
 Author's personal copy
Multimedia Tools and Applications (2020) 79:13125–13138
https://doi.org/10.1007/s11042-020-08652-9

On the implementation of a copyright protection

scheme using digital image watermarking

Subhrajit Sinha Roy 1,2 1

& Abhishek Basu & Avik Chattopadhyay
2

Received: 18 January 2019 / Revised: 5 November 2019 / Accepted: 3 January 2020 /

Published online: 27 January 2020
# Springer Science+Business Media, LLC, part of Springer Nature 2020

Abstract
In this paper, a spatial domain based digital image watermarking scheme has been developed
to serve the purpose of copyright protection for digital images. This scheme proposes how to
embed copyright information, i.e. the watermark bits by adaptive LSB replacement tech-
nique. Here, to improve imperceptibility, imperfect nature of human visual system has been
exploited by means of saliency map generation. Robustness and data hiding capacity are also
optimized by modifying the higher bit-planes instead of only the LSB. A few signal
processing attacks are applied to evaluate the robustness, and several image quality metrics
are involved in this purpose. The quality metrics are also used in verifying experimental
results in terms of imperceptibility; and finally the system proficiency is compared to some
existing frameworks. A proper benchmarking is also employed for a quantitative justifica-
tion of the visual image degradation, caused by watermark insertion.

Keywords Adaptive . Imperceptible . LSB . Payload . Robust . Saliency . Spatial

1 Introduction

Now-a-days, lots of multimedia objects are exchanged over network every time, and data
augmentation has become easier in digital domain. Therefore, the copyright infringement has
turned into a vital issue. Among a lot of information hiding tools [25, 42], the practice of
digital watermarking has become the most trendy in serving the purpose of copyright
protection. Besides, cryptography and steganography [1–5, 17, 41], two of the most conven-
tional arts, are often used till date for information hiding as well as for secret transmission. But,

* Subhrajit Sinha Roy

[email protected]

1
Electronics and Communication Engineering Department, RCC Institute of Information Technology,
South Canal Road, Beliaghata, Kolkata 700015, India
2
Institute of Radio Physics and Electronics, University of Calcutta, 92 Acharya Prafulla Chandra Road,
Kolkata 700009, India
 Author's personal copy
13126 Multimedia Tools and Applications (2020) 79:13125–13138

digital watermarking offers several advantages over them [46]. For instance, in case of digital
watermarking, the hidden object, carrying the copyright information, is made robust at top
priority, one-to-many communication can be achieved, no secret transmission is required, etc.
That is why digital watermarking has become the mostly accepted tool amongst the researchers
in this field.
Digital watermarking is the process of embedding information or mark into a multimedia
object, such that the mark can be extracted or detected at the receiving end. Till yet, for any
type of multimedia objects, the information insertion process has been performed mainly in
two domains. One rolls up in spatial domain, where selected pixels are directly modified [9].
The other one is the frequency domain, where the watermarking processes are basically
developed using discrete cosine transform (DCT) [40], discrete Fourier transform (DFT)
[31], discrete wavelet transform (DWT) [43], etc. Here, secret data is embedded into the best
frequency portions of the protected image. Multiple transforms [57] may also be utilized to
improve system performance. Frequency domain offers a better robustness, but spatial domain
schemes are simple to implement, less time consuming, and offering a better imperceptibility
with respect to the frequency domain methods. That is why this paper aspires to develop a
watermarking scheme in spatial domain.
The acceptability of any digital watermarking algorithm is primarily evaluated by three
qualitative parameters– imperceptibility, robustness, and payload (hiding capacity). It has been
found in the surveys on digital watermarking [28, 35, 36] that, the three aforesaid parameters
oppose each other while trying to improve them together. To deal with this issue, as a recent
trade, utilization of the imperfect nature of human visual system (HVS) has been introduced
[13]. In this practice, salient object detection by means of saliency map generation [6, 49],
visual attention model (VAM) [7], just noticeable difference (JND) [32, 45] are involved. The
general aim of these methods is to indicate the locations in cover image, where the watermark
signal, with utmost strength, can be embedded without any major noticeable distortion. To
compute VAM, there exist many techniques, which emulate the pattern of the HVS in selecting
the regions of interest (ROI). Several watermarking algorithms have been developed depend-
ing on these types of visual models. In 1998, Hung and Shi [23] developed a spread spectrum
based image watermarking scheme that utilizes human visual perception. Here, brightness and
texture sensitivity were considered in DCT domain, and the watermark sequences were
inserted into low frequency coefficients. Kim et al. [26] used wavelet transform instead of
DCT. In this approach, the changing rate of a sinusoidal pattern per subtended visual angle was
considered as the estimation of visual weight of watermark in each wavelet band. In 2005, Lee
et al. [30] introduced a ROI-based medical image watermarking. This was a non-blind
approach, as the interested portions of the watermark were embedded into the regions of
noninterest. This scheme was robust against JPEG compression impairments only. In 2006,
Chu et al. [10] inserted watermark in the interior of ROI. Que. et al. [44] improved robustness
against image patching and cropping with the help of HVS based watermarking scheme.
Another ROI dependent watermarking was proposed by Ni and Ruan [38], and Fan et al. [12]
where, watermark was embedded into ROI (selected by the user) of the image. Instead of
selecting ROI areas by user, Mohanty et al. [37] used various influencing features of HVS like
intensity, contrast, edginess, texture and location in their watermarking scheme. Basu et al. [6]
proposed an adaptive spatial domain image watermarking technique based on bottom-up
graph-based visual saliency. It embeds watermark to the uneven bit-depth salient image pixels,
and provides better perceptual transparency as well as an improved robustness. Another
saliency based watermarking technique was described by Niu et al. [39]. It can be used as a
 Author's personal copy
Multimedia Tools and Applications (2020) 79:13125–13138 13127

guide to optimize image watermarking by producing a visual saliency modulated JND profile.
Apart from the psychosomatic or mental status of HVS, a simple pixel value based spatial
domain saliency map was generated by A. Basu et al. [8] in 2016. Here, the pixels, with higher
dissimilarities in value according to the neighbor pixels, were considered as higher salient
regions. Being utterly developed in spatial domain, the computation time and complexity were
reduced in a remarkable rate.
The proposed watermarking method has been developed in spatial domain. Here, Spectral
Residual based saliency model is followed to estimate a hiding capacity map, depending on
which, watermark is implanted within the cover object through an adaptive LSB replacement
technique. Section-2 consists of the explanation of spectral residual based saliency map. In
section-3 the watermark insertion and extraction procedure is elaborated along with a suitable
example. The results of this projected scheme are discussed and compared to some other
existing schemes in section-4. Finally, this paper is concluded in section-5.

2 Saliency map model

Saliency is the feature of an object, through which it can stand out over the neighboring others.
The purpose of introducing saliency in watermarking is to enhance the imperceptibility by
utilizing the imperfect nature of HVS. Actually, the HVS does not pay equal attention to the
whole of any visual stimuli; rather, it is more sensitive to the striking or salient regions.
Saliency detection [11, 14, 52] is a complex nervous and psychological decision of human
beings, and often, it depends upon the mental status also. Many interacting factors like bottom-
up and top-down factors contribute to make this decision. The former are all those depend only
on the instantaneous sensory input, without taking into account the internal state of the
organism. Top-down control, on the other hand, takes account into the internal state. A
bottom-up saliency map approach is proposed by Itti et al. [24]. Some other saliency detection
algorithms like GBVS [19], spectral residual based saliency [22], simpsal, signature saliency
[21] are also developed during the last decades. In spite of the presence of more recent
techniques, the Spectral Residual Model has been used in this proposed watermarking method
for several advantages [22]. This is a generalized model that does not depend on the prior
knowledge of the image.
This method is based on the perspective of information theory. Through proper coding, any
image signal can be decomposed into two parts–innovation part (indicating the novelty part)
and prior knowledge (indicating the redundancy that ought to be suppressed by some useful
coding). Therefore, the saliency map could be generated by approximating the novelty part of
an image by eliminating the statistical redundant components. This may be achieved using
spectral residual method in an easy and effective way.
Firstly, the discrete Fourier transform of the image under experiment is obtained to get the
frequency coefficients of the image pixels, and from that Fourier spectrum, the amplitude and
phase, related to each pixel of that image are worked out. Next, the log spectra of the amplitude
curve are estimated within the range of 0 to 1, and convoluted with an espoused local average
filter [22, 46] to approximate the regular shape of log spectra. This carries the prior information
of the concerned image. According to Hou et al.’s approach, the statistical singularities,
pouncing from this general curve, are exacting the innovation part that is relevant to that
particular image only, and denoted as the spectral residual of that image. Finally, through
inverse Fourier transform, the saliency map is originated in spatial domain, and a Gaussian
 Author's personal copy
13128 Multimedia Tools and Applications (2020) 79:13125–13138

Cover Image Adaptive LSB Substitution Watermarked

(Gray) Image
Saliency Hiding Capacity Map
Map
Watermark
(Binary)

Fig. 1 Block Diagram for Watermark Insertion

filter is introduced for smoothing purpose. In this manner, the overall saliency map for any
given image is formulated using spectral residual model.

3 Watermark insertion and extraction

The proposed algorithm, which deals with the trade-off between robustness and
imperceptibility, utilizes the perceptual characteristics of HVS during the watermark insertion
process to provide high data transparency. In this issue, first a saliency map is obtained through
which a hiding capacity map is formed. Then an adaptive LSB replacement is performed to the
cover image to develop the watermarked image. Figures 1 and 2 show the block diagrams of
watermark embedding and extracting systems, respectively, specifying the steps, involved in
producing the watermarked image.

3.1 Watermark insertion

A grayscale image O of size (M × N) has been chosen as a cover or host image and as a
watermark image, a binary image T of size (X × Y) is taken. Therefore, O and T can be defined
as,
O ¼ foðm; nÞj1≤m ≤M ; 1≤n ≤N ∧oðm; nÞϵ ½0; 1; 2; 3; …255g ð1Þ

T ¼ ftðx; yÞj1≤x ≤X ; 1≤y ≤Y ∧tðx; yÞε½0; 1g ð2Þ

Step 1: Evaluating the saliency map from the host image

The saliency map of the cover image O is obtained through spectral residual method as
described in section 2. The saliency map vector V can be defined as,

Watermarked Adaptive LSB Substitution Similitude

Image consideration for
(Gray) Saliency retrieved bits
Hiding Capacity Map
Map

Recovered
Watermark

Fig. 2 Block Diagram for Watermark Extraction

 Author's personal copy
Multimedia Tools and Applications (2020) 79:13125–13138 13129

V ¼ fvðm; nÞj1≤m ≤M ; 1≤n≤N ; 0≤vðm; nÞ ≤1∧vðm; nÞϵRg ð3Þ

Using eq. (3), the grade of saliency for each pixel in O can be defined within the range of 0 to 1
through the analogous pixel value in V. The 0 values in V indicate lowest saliency regions, and
1 represents the pixels with highest saliency.
Step 2: Constructing the Hiding Capacity Map from the Saliency Map

The proposed watermarking scheme basically works on the concept of adaptive LSB replace-
ment technique that embeds watermark bits by replacing one or more LSBs from each of the
pixels of the cover image. Here, multiple numbers of bits are replaced instead of a single bit,
i.e. the LSB, intending to provide improved robustness. It is because the effects of most of the
attacks gradually lessen from LSB plane to MSB plane. Moreover, it offers increased payload.
In saliency map, the pixel values from 0 to 1 indicate the nature of the corresponding pixel in
the cover image as flat region to busy region. The ability of HVS to detect any slightest change
is lesser in the busy region than that in a flat region. Therefore, the busy region pixels have
been chosen to hide maximum amount of watermark bits to maintain the visual aesthetic.
Assuming that the maximum number of LSBs could be replaced in each pixel of the busiest
region of the cover is chosen as K, and for the flat region pixels a single bit i.e. only LSB is to
be replaced by the watermark bits. Determination of the value of K is a process of
experimenting with selection of different values of bits (from 1 to 8), so that the conflict
between imperceptibility and payload could be optimized. Now, if the number of K is
decreased by 1 with decreasing of saliency, the cover image pixels should be divided into K
number of regions. In this purpose, a hiding capacity map (HCM) has been developed, through
which the cover image pixels are sectored into K distinct regions based on their corresponding
saliency map values. HCM generation is completely performed by fixing up several ranges of
the saliency map values. Therefore, the HCM, denoted as H, can be considered as a function of
the saliency map V.
Now, if v(m,n) ∈ V and h(m,n) ∈ H, then a function fi: V ➔ H is defined as,

hðm; nÞ ¼ 2ð7−K Þ ∑iðb¼1Þ 2b for yði−1Þ:10i–K–1 ≤vðm; nÞ ≤yðiÞ:10i–K ð4Þ

where, i ∈ [1, 2… K], is the number of LSBs chosen to be replaced from the particular cover
image pixel o(m,n) concerned to v(m,n) and y(i) is defined as,
yðiÞ ¼ 1 for 1≤i ≤K
¼ 0 otherwise

Therefore,
H ¼ fhðm; nÞj1≤m ≤M ; 1≤n≤N ∧hðm; nÞ ¼ f i ðvðm; nÞÞg ð5Þ

Step 3: Constituting the watermarked image through adaptive LSB substitution

The last stage of the embedding system is adaptive LSB substitution, which is performed
depending on the HCM pixel values. As already mentioned, K is the maximum number of
LSBs, to be replaced for the cover pixels, providing maximum hiding capacity in HCM. With
the diminution of hiding capacity, lesser numbers of LSBs are opted to be replaced for
 Author's personal copy
13130 Multimedia Tools and Applications (2020) 79:13125–13138

corresponding cover image pixels. This algorithm also enables multiple embedding of the
watermark. In this quest, the size of watermark is matched with the cover size by repeating the
mark pattern within the size interval of cover image. The new watermark TM,N consists of
multiple number of original watermark T can be constructed as,
2 3
T T ⋯ T
6 T T ⋯ T 7
T M ;N ¼ 64⋮ ⋮ ⋮ ⋮5
7

T T ⋯ T M N :

Therefore, the new watermark TM,N can be defined as,

T M ;N ¼ ftðm; nÞj1≤m ≤M ; 1≤ n≤N ∧tðm; nÞ∈½0; 1g ð6Þ

If Q is the watermarked image, q(m,n) ∈ Q and t(m,n) ∈ TM,N then the function fE: O × H ×
TM,N → Q is defined as,

qðm; nÞ ¼ ∑7b¼i ob ðm; nÞ2b þ ∑i−1

b¼0 t ðm; nÞ2
b
for hðm; nÞ ¼ 27−K ∑ib¼1 2b ð7Þ

where, ob(m, n) is the bth bit (considering LSB as the 1st bit) of the pixel o(m,n). Therefore, the
watermarked image Q is finally defined as,
Q ¼ fqðm; nÞj1≤m ≤M ; 1≤n≤N ∧qðm; nÞ ¼ f E ðoðm; nÞ; hðm; nÞ; tðm; nÞÞg ð8Þ

From eq. (8) it is reflected that the watermarked image Q is a gray scale image with the size
same as the cover image O.

3.2 Watermark extraction

The purpose of watermark extracting is to recover the watermark from a received or

watermarked image so that the originality could be evaluated by comparing it to the original
mark. While watermark extracting process, same algorithm is followed to generate the saliency
map V R and the hiding capacity map H R for the received image. Data points in H R are used to
obtain the number of watermark bits, embedded in a particular pixel of the cover image. For
any watermarked image pixel, if the corresponding HCM map value is (27−K ∑ib¼1 2b ), then it
indicates that i numbers of bits are inserted in that particular pixel. Therefore, from that pixel, i
numbers of bits (starting from 1st LSB) should be extracted. Finally, through a similarity
estimation of the extracted bits from each pixel, the actual watermark bit is approximated. The
similarity estimation is essential for reducing the error probability, as bit-planes may be
affected by the signal processing attacks. If S is considered as the set of the extracted bits
from any particular pixel, the approximated watermark bit tr is set to 0 or 1 through a function
fs, that generates the maximum occurred bit in the set S using eq. (9).
tr ¼ s1 when; S ¼ ½s1 
¼ s2 when; S ¼ ½s1 ; s2 
ð9Þ
¼ ðs1 AND s2 ÞORðs2 AND s3 ÞORðs3 AND s1 Þ when; S ¼ ½s1 ; s2 ; s3 
¼ ðs4 AND ðs1 OR s2 OR s3 ÞÞORðs1 AND s2 AND s3 Þ when; S ¼ ½s1 ; s2 ; s3 ; s4 
where, s1, s2, s3, and s4 indicate the 1st, 2nd, 3rd, and 4th LSBs, respectively extracted from
any particular watermarked image pixel.
 Author's personal copy
Multimedia Tools and Applications (2020) 79:13125–13138 13131

Here, fs is developed to perform similarity estimation for maximum number of four bits, as
the value of K, i.e. the maximum number of bits, to be replaced from the busiest region, is
chosen as 4 in this paper.
Now, it is obvious that the construction of B for each pixel of QR depends on that pixel
value along with its corresponding HCM value in HR. Therefore, the recovered watermark is
finally obtained as a function of QR and HR and can be stated as,
T R ¼ ft r ðx; yÞj1≤x≤ X ; 1 ≤y≤Y ∧t r ðx; yÞ ¼ f s ðSÞwhere S ¼ f ðQR ; H R Þg ð10Þ

3.3 Example to demonstrate for watermark insertion and extraction

In this section, the watermark embedding and extracting processes are realized using two test-
matrixes, taken as the watermark T and cover O. TM,N is formed by repeating the pattern of T,
as described previously.
2 3 2 3
  127 43 0 0 0 1 0 1
0 1 6 127 43 255 0 7 6 1 0 1 07
T¼ O¼4 T M ;N ¼ 4
1 0 127 43 43 43 5 0 1 0 15
127 127 127 127 1 0 1 0

In next step, through spectral residual method, the saliency map V is generated for O. The
hiding capacity map H is formed for O using eq. (4), based on the data-set values of V. Here,
the HCM consists of four distinct values, as K, the maximum number of LSBs to be replaced
from the most salient region, is chosen as 4.
2 3 2 3
0:2012 0:0006 0:0196 0:0276 240 16 112 112
6 0:2471 0:0019 1 0:0196 7 6 240 48 240 112 7
V ¼4 H ¼4
0:1963 0:0001 0:0019 0:0006 5 240 16 48 16 5
0:2104 0:1963 0:2471 0:2012 240 240 240 240

It is clear from the matrixes V and H that the hiding capacity for pixels in O increases with
increment of saliency map values for the corresponding pixels. Therefore, according to the
proposed scheme, first four, three, two, and one LSB(s) of the cover image pixels are to be
replaced by instantaneous watermark bit, when the analogous HCM values are found as 240, 112,
48, and 16, respectively. Thus, the watermarked image matrix Q is formed, as described in eq. (7).
2 3
112 43 0 7
6 127 40 255 0 7
Q¼4
112 43 40 43 5
127 112 127 112

In case of extracting the watermark from received image QR (here considered as Q), the
saliency map V R and HCM H R are generated for Q in the same way, as followed in embedding
process. Here, it is observed that the data points of V R has been changed with respect to V, as
pixel values in Q differs from those in O; but, H R is identical to H. This occurs, because the
relative saliency for each of the cover image pixels does not change, and thus, it belongs within
the same range, as prescribed in eq. (4), on which the corresponding data value in HCM
depends.
 Author's personal copy
13132 Multimedia Tools and Applications (2020) 79:13125–13138

 
0:1752
0:0002 0:0145 0:0229 0:1757 0:0001 0:0041 0:0002
VR ¼ HR
0:3228
0:0041 1 0:0145 0:2976 0:1757 0:3228 0:1752
 
240 16 112 112 240 16 48 16
¼
240 48 240 112 240 240 240 240

As mentioned in the watermark extraction process, first four, three, two, and one LSB(s) are
extracted from the pixels of Q, having corresponding HCM values as 240, 112, 48, and 16. In
this way, S, i.e. the set of extracted LSBs for each pixel in Q is generated. To envisage these all
sets together, a matrix E is formed here through replacing each data-point in matrix Q by its
consequent S. Next, the similarity estimation is performed for each S in E, abiding by eq. (9);
and it results in developing another matrixE, as given bellow. With a prior information on the
actual size of the original mark, watermark TR can be constructed from E.
2 3 2 3
½0 0 0 0 ½1 ½0 0 0 ½1 1 1 0 1 0 1  
6 ½1 1 1 1 ½0 0  ½1 1 1 1 ½0 0 0 7 61 0 1 07 0 1
E¼4 E¼4 TR ¼
½0 0 0 0 ½1 ½0 0 ½1 5 0 1 0 1 5
1 0
½1 1 1 1 ½0 0 0 0 ½1 1 1 1 ½0 0 0 0 1 0 1 0

4 Results and discussion

The USC-SIPI image database [50] has been used as a set of host or cover images after
modifying each image to a 256 × 256 Gray-scale image. And as the watermark image, a binary
image of size (16 × 16) (Fig. 3a) is built up. The output results for some of the cover images,
obtained as Saliency Map, Hiding Capacity Map and watermarked images are shown in Fig.
3b–e, respectively.
In Fig. 3, it is seen that the saliency maps highlighted the most salient or busiest regions
where, any type of small changes cannot be optimized by HVS. In hiding capacity map, the
brightest area indicates the most preferable regions for data embedding and lesser bright
regions indicates regions having lower hiding capacity respectively.
An important feature of a good watermarking scheme is Data Hiding Capacity, measured in
bits-per-pixel (bpp) scale. It defines as the average number of bits of the watermark can be
embedded in a single pixel of the host image. In the proposed algorithm, the average hiding
capacity is estimated as 2.80 bits/pixel. From HCM the value can be determined as it is already
stated in eq. (4) that for a pixel in HCM having a value of h(m,n), i number of bits can be
embedded. From this concept, total number of embedded bits is calculated. Now hiding
capacity is measured as the ratio of the total number of bits embedded into an image to the
total number of pixels.

Fig. 3 a Watermark image (16 × 16); b Original images; c Saliency map of the original images; d HCM for the
original images; e Watermarked images
 Author's personal copy
Multimedia Tools and Applications (2020) 79:13125–13138 13133

The visual perception measurement is achieved through some image quality metrics [29,
47] to differentiate between original image and watermarked image. These perceptual qualities
together indicate imperceptibility of the proposed algorithm. From Table 1 it is reflected that
this algorithm provides increased PSNR typically has an average value about 50 dB whereas it
is measured maximum 68.32 dB for the image 5.1.13.tiff. The maximum difference is 15 for
all the images as the number of maximum LSBs replaced in a pixel is 4. Image fidelity (IF)
should be 1 for two identical images. Here, it is found around 0.9999, which is very close to
unity. Middling value of Structural Content (SC) and Normalized Cross-Correlation (NCC) for
most the images are more than 0.999, which is also 1 for two indistinguishable images.

Table 1 Performance outcome in terms of imperceptibility

Image MSE PSNR SNR MD AD NAD SSIM IF SC NCC

4.1.01 0.445 50.90 40.54 15 0.0905 0.00154 0.999379 0.99991 0.99808 0.99908
4.1.02 0.448 51.24 36.74 15 0.0912 0.00273 0.999670 0.99979 0.99850 0.99936
4.1.03 0.198 55.16 50.12 7 0.0583 0.00042 0.999924 0.99999 0.99945 0.99973
4.1.04 0.202 54.90 48.65 15 0.0595 0.00054 0.999789 0.99999 0.99928 0.99964
4.1.05 0.326 52.43 48.12 8 0.0779 0.00056 0.999531 0.99998 0.99951 0.99976
4.1.06 0.652 49.36 45.13 15 0.1001 0.00078 0.999621 0.99997 0.99923 0.99963
4.1.07 0.165 54.02 52.93 7 0.0544 0.00031 0.999609 0.99999 0.99965 0.99983
4.1.08 0.206 53.21 51.60 13 0.0594 0.00036 0.999537 0.99999 0.99963 0.99982
4.2.01 0.577 50.14 43.62 15 0.1092 0.00106 0.999144 0.99996 0.99945 0.99974
4.2.02 0.436 51.73 50.17 15 0.0898 0.00043 0.999352 0.99999 0.99967 0.99984
4.2.03 2.385 42.79 38.84 15 0.2317 0.00179 0.999594 0.99987 0.99794 0.99903
4.2.04 0.471 50.88 45.74 15 0.0861 0.00069 0.999582 0.99997 0.99898 0.99950
4.2.05 0.306 52.45 50.48 7 0.0746 0.00042 0.999683 0.99999 0.99959 0.99980
4.2.06 1.409 46.15 41.50 15 0.1644 0.00131 0.999433 0.99993 0.99872 0.99940
4.2.07 0.670 48.82 44.12 15 0.1087 0.00090 0.999683 0.99996 0.99904 0.99954
5.1.09 1.069 47.63 42.04 15 0.1436 0.00112 0.999112 0.99994 0.99877 0.99941
5.1.10 1.664 45.68 41.17 15 0.1846 0.00131 0.999399 0.99992 0.99869 0.99938
5.1.11 0.699 48.86 47.41 15 0.1203 0.00062 0.998898 0.99998 0.99963 0.99983
5.1.12 0.193 54.97 52.93 15 0.0581 0.00031 0.999617 0.99999 0.99961 0.99981
5.1.13 0.010 68.32 67.73 3 0.0075 0.00003 0.999964 1.00000 0.99993 0.99997
5.1.14 0.666 49.89 42.80 15 0.1156 0.00111 0.999690 0.99995 0.99855 0.99930
5.2.08 1.425 46.59 40.67 15 0.1529 0.00124 0.999921 0.99991 0.99893 0.99951
5.2.09 1.539 46.26 43.44 15 0.1766 0.00098 0.999583 0.99995 0.99902 0.99953
5.2.10 2.185 44.74 38.58 15 0.2195 0.00193 0.999453 0.99986 0.99776 0.99895
5.3.01 0.657 49.17 42.29 15 0.1095 0.00123 0.999769 0.99994 0.99888 0.99947
5.3.02 0.983 47.57 39.05 15 0.1345 0.00162 0.999695 0.99988 0.99843 0.99928
7.1.01 1.669 42.59 38.63 15 0.1827 0.00171 0.999047 0.99986 0.99849 0.99931
7.1.02 0.228 52.77 51.36 15 0.0614 0.00035 0.999445 0.99999 0.99954 0.99977
7.1.03 0.690 47.55 44.22 15 0.1196 0.00090 0.999015 0.99996 0.99891 0.99947
7.1.04 1.158 45.89 41.03 15 0.1508 0.00130 0.999130 0.99992 0.99883 0.99946
7.1.05 0.707 46.85 42.46 15 0.1197 0.00113 0.999609 0.99994 0.99882 0.99944
7.1.06 1.126 44.19 39.13 15 0.1475 0.00163 0.999390 0.99988 0.99857 0.99935
7.1.07 0.809 45.53 41.80 15 0.1262 0.00117 0.999387 0.99993 0.99873 0.99940
7.1.08 0.477 49.06 45.45 15 0.0951 0.00075 0.999482 0.99997 0.99922 0.99962
7.1.09 0.693 47.79 43.90 15 0.1188 0.00095 0.999471 0.99996 0.99905 0.99955
7.1.10 1.378 43.76 40.34 15 0.1656 0.00139 0.998972 0.99991 0.99863 0.99936
7.2.01 0.199 52.17 39.27 15 0.0568 0.00175 0.999520 0.99988 0.99811 0.99911
boat.512 0.376 52.10 47.01 15 0.0758 0.00058 0.999585 0.99998 0.99944 0.99973
ealine.512 0.832 48.51 43.95 15 0.1236 0.00091 0.999445 0.99996 0.99926 0.99965
gray21.512 0.332 52.92 48.22 15 0.0527 0.00042 0.999074 0.99998 0.99957 0.99979
numbers.512 1.328 46.90 40.10 15 0.1596 0.00154 0.999669 0.99990 0.99859 0.99934
ruler.512 0.320 53.09 52.18 15 0.0552 0.00025 0.999901 0.99999 0.99986 0.99993
testpat.1 k 0.310 53.21 48.04 7 0.0757 0.00061 0.999842 0.99998 0.99936 0.99969
 Author's personal copy
13134 Multimedia Tools and Applications (2020) 79:13125–13138

Table 2 Performance outcome in terms of robustness

Sl. No Attacks BER (%) PCC SM WDR WPSNR

1. No attack 0 1 1 –∞ ∞
2. 180° Rotation 0 1 1 –∞ ∞
3. 90° Rotation 0 1 1 –∞ ∞
4. 45° Rotation 1 0.9539 0.999989 −39.1202 31.2620
5. 20° Rotation 3.5 0.8979 0.999983 −31.0109 26.4242
6. Negative 0 1 1 –∞ ∞
7. LSB Inversion (1st LSB ↔ 2ND LSB) 0 1 1 –∞ ∞
8. Cropping 0 1 1 –∞ ∞
9. Erode 2.7 0.9239 0.999994 −33.5241 30.5690
10. Salt & Pepper 0 1 1 –∞ ∞
11. Median Filtering 0.4 0.9887 1 −52.9832 40.6245
12. Dilate 2.3 0.9310 0.999978 −35.0656 28.2749
13. Scaling 24 0.4121 0.995234 −11.8744 17.0655
14. JPEG Compression 36.72 0.0263 0.970413 −7.5502 12.2050

Therefore, this approach can be considered as a good imperceptible algorithm which also
reflects from the values of the other parameters.
Table 2 exhibits the robustness of the proposed method against the above said spiteful
attacks in comprehensive forms. It bears out that this proposed method is robust against most
of the attacks. This scheme is least robust against Scaling and JPEG compression attacks, as
these attacks basically distort the LSB plane and where the watermark bits are embedded
according to our proposed method. Although rotation attacks are resisted in this method, it is
moderately robust against non-uniform rotations. As for rotational and negative attacks the
changes are visualized in aesthetic sense, a reverse process corresponding to the attack is
applied to the noisy images before extracting the watermark. Also for dilate attack this scheme
just offers an average sturdiness. Except from these, watermark can excellently persist against
other attacks with 0% bit error rate as described here.
A relative study between the proposed algorithm and some other existing methods is shown
in Table 3. It is obvious that the watermark and cover images, used in different proposals, are

Table 3 Results of proficiency comparison in terms of imperceptibility and payload capacity

Sl. No. Method PSNR (dB) Maximum Payload (bpp)

1. Proposed Method 49.74 2.80

2. Salient Region Watermarking [53] 45.83 0.58
3. Pair wise LSB matching [55] 35.05 2.25
4. Optimal LSB pixel adjustment [56] 34.84 4
5. Matrix Encoding based Watermarking [51] 55.91 1.99
6. LSB Replacement method [15] 54.8 1
7. DWT and SVD based Watermarking [33] 41.72 0.375
8. Adaptive Pixel Pair Matching Scheme [20] 40.97 3
9. Reversible Data Hiding Scheme [16] 34.26 1
10. Magic Cube based Information Hiding [54] 45.15 2
11. Integer Wavelet and SVD based watermarking [34] 44.00 2
12. Spatial Domain based Watermarking [48] 49.99 0.0013
13. Wavelet based watermarking through SPIHT [27] 36.77 3.5
14. Robust Color Image Watermarking [18] 50.00 0.02
 Author's personal copy
Multimedia Tools and Applications (2020) 79:13125–13138 13135

Table 4 Results of robustness comparison by means of NC

Sl. Method Rotation Salt & Median Scaling JPEG

No. Attack Pepper Filtering Attack Compression
Attack Attack Attack

1. Proposed Method 0.979 0.995 1 0.789 0.795

2. Integer Wavelet and SVD based 0.975 0.995 0.989 0.984 0.998
watermarking [34]
3. Spatial Domain based 0.778 0.997 0.970 0.998 0.734
Watermarking [48]
4. Wavelet based watermarking 0.998 0.896 0.953 0.556 0.997
through SPIHT [27]
5. Robust Color Image 0.868 0.958 0.976 0.990 0.990
Watermarking [18]

not identical in size or type. An average value through watermark to document scale is
approximated. For the proposed method, the values of PSNR and maximum payload, given
in Table 3, are estimated by averaging over all their corresponding values, obtained for the
images of USC-SIPI image-database. The payload for this projected scheme is provided as the
average maximum hiding capacity, i.e. maximum amount of information could be embedded;
and it does not mean that this amount of data is embedded for all times. Here, it is observed
that this proposed watermarking technique offers much better PSNR and a remarkable data
hiding capacity than most of the schemes. Most of the methods are inferior with respect to the
offered algorithm in terms of imperceptibility and capacity. Only the Matrix encoding based
and LSB replacement based methods bid a close PSNR, whereas, their payload capacities are
reduced. In a contrast, IP LSB and Optimal LSB adjustment techniques provides improved
payload but their low PSNR values confirm that these methods are not so good in terms of
imperceptibility.
The robustness is compared by means of normalized cross-correlation (NC) after different
attacks. In Table 4 it is observed that in terms of robustness, this proposed scheme can compete
well with some recent robust image watermarking methodologies too.
A quantitative description about the image degradation caused by the watermarking process
can be given through a proper benchmark. Here ITU-R Rec. 500 quality rating [29] on a scale
from 1 to 5 is used to investigate the benchmark. Under this benchmark condition this
proposed watermarking scheme is rated as 4.71 which is very high and can be considered as
an excellent rating according to ITU-R Rec. 500 quality rating. The rating obtained for
different images under experiment is shown in Fig. 4.

5.00
Benchmark Rating

4.90
4.80
4.70
4.60
4.50
4.40
4.30
4.20
4.10
elaine.512

testpat.1k
boat.512
4.1.01
4.1.02
4.1.03
4.1.04
4.1.05
4.1.06
4.1.07
4.1.08
4.2.01
4.2.02
4.2.03
4.2.04
4.2.05
4.2.06
4.2.07
5.1.09
5.1.10
5.1.11
5.1.12
5.1.13
5.1.14
5.2.08
5.2.09
5.2.10
5.3.01
5.3.02
7.1.01
7.1.02
7.1.03
7.1.04
7.1.05
7.1.06
7.1.07
7.1.08
7.1.09
7.1.10
7.2.01

gray21.512
numbers.512
ruler.512

Fig. 4 Quality rating for different images

 Author's personal copy
13136 Multimedia Tools and Applications (2020) 79:13125–13138

5 Conclusion

A spatial domain based image watermarking technique has been proposed here, that imper-
ceptibly embeds copyright information into cover image using adaptive LSB replacement
process. HVS analysis has been introduced through saliency detection for obtaining better
visual transparency with increased payload capacity. A comparison with some other existing
frameworks ascertains this proposed scheme as a superior invisible image watermarking
technique having enhanced data capacity. Even under the environment, prone to several kinds
of attacks, this method has confirmed an improved robustness too. Therefore, in a brief, it can
be concluded that the watermarking scheme, developed in this work, is able to optimize the
conflicts between imperceptibility and data hiding capacity, besides, providing some notewor-
thy robustness. In this context, it is worth mentioning that this proposed scheme also has a very
good prospect for hardware implementation, besides, the opportunity of improvement in
system performance with the melioration of HVS model. Moreover, this scheme can be
adapted to embed copyright information into color images as well as the other multimedia
objects, like text, audio, and video.

References

1. Abu-Marie W, Gutub A, Abu-Mansour H (2010) Image based steganography using truth table based and
determinate Array on RGB Indicator. International Journal of Signal and Image Processing 1(3):196–204
2. Alanazi N, Alanizy A, Baghoza N, Al Ghamdi M, Gutub A (2018) 3-layer PC text security via combining
compression, AES cryptography 2LSB image steganography. Journal of Research in Engineering and
Applied Sciences 3(4):118–124
3. Al-Ghamdi M, Al-Ghamdi M, Gutub A (2018) Security enhancement of shares generation process for
multimedia counting-based secret-sharing technique. Multimed Tools Appl, Springer 78(12):16283–16310
4. Al-Juaid NA, Gutub AA, Khan EA (2018) Enhancing PC Data Security via Combining RSA Cryptography
and Video Based Steganography. Journal of Information Security and Cybercrimes Research 1(1)
5. Alsaidi A, Al-Lehaibi K, Alzahrani H, Al Ghamdi M, Gutub A (2018) Compression multi-level crypto
Stego security of texts utilizing colored email forwarding. Journal of Computer Science and Computational
Mathematics 8(3):33–42
6. Basu, A., Das, T.S., Sarkar, S.K., 2011, “On the implementation of a information hiding design based on
saliency map”, Int Conf Image Information Processing, IEEE, p. 1–6.
7. Basu A, Sarkar SK (2013) “On the implementation of robust copyright protection scheme using visual
attention model”, information security journal: A global perspective. Taylor & Francis 22(1):10–20
8. Basu A, Sinha Roy S, Chattopadhayay A (2016) Implementation of a spatial domain salient region based
digital image watermarking scheme”, Int. Conf. Research in computational intelligence and communication
networks, IEEE, p. 269-272
9. Chopra D, Purohit R, Gaur SBC (2014) Quality evaluation for LSB replacement watermarking of Grayscale
images. Int Journal of Advanced Research in Computer and Communication Engineering 3(5)
10. Chu Y, Zhang Y, Zhang S, Ye X (2006) Region of interest fragile watermarking for image authentication.
First International Multi-Symposiums on Computer and Computational Sciences 1:726–731
11. Duan L, Wu C, Miao J, Qing L (2011) Visual saliency detection by spatially weighted dissimilarity. IEEE
Conference on Computer Vision and Pattern Recognition:473–480
12. Fan YC, Chiang A, Shen JH (2008) ROI-based watermarking scheme for JPEG 2000. Circuits, Systems &
Signal Processing 27(5):763–774
13. Foris P, Levicky D (2007) Implementations of HVS models in digital image watermarking. Radio
Engineering 16(1):45–50
14. Goferman S, Zelnik-Manor L, Tal A (2013) Context-aware saliency detection. IEEE Trans Pattern Anal
Mach Intell 34(10):1915–1926
15. Goyal R, Kumar N (2014) LSB based digital watermarking technique. International Journal of Application
or Innovation in Engineering & Management 3(9):15–18
 Author's personal copy
Multimedia Tools and Applications (2020) 79:13125–13138 13137

16. Gui X, Li X, Yang B (2014) A high capacity reversible data hiding scheme based on generalized prediction-
error expansion and adaptive embedding. Signal Process 98:370–380
17. Gutub A, Al-Ghamdi M (2019) Image based steganography to facilitate improving counting-based secret
sharing. 3D Res, Springer 10(1)
18. Han S, Yang J, Wang R, Jia G (2018) A robust color image watermarking algorithm against rotation attacks.
Optoelectron Lett 14(1):61–66
19. Harel J, Koch C, Perona P (2006) Graph-based visual saliency. Proc Neural Information Processing
Systems:545–552
20. Hong W, Chen TS (2012) A novel data embedding method using adaptive pixel pair matching. IEEE
Transactions on Information Forensics And Security 7(1):176–184
21. Hou X, Harel J, Koch C (2012) Image signature: highlighting sparse salient regions. IEEE Trans Pattern
Anal Mach Intell 34(1):194–201
22. Hou X, Zhang L (2007) Saliency detection: A spectral residual approach. Int Conf Computer Vision and
Pattern Recognition, IEEE:1–8
23. Huang J, Shi YQ (1998) Adaptive image watermarking scheme based on visual masking. IEE Electronics
Letters 34(8):748–750
24. Itti L, Koch C (2000) A saliency-based search mechanism for overt and covert shifts of visual attention. Vis
Res 40(10–12):1489–1506
25. Katzenbeisser S, Petitcolas FAP (2000) Information hiding techniques for steganography and digital
watermarking. Artech House, Boston, London
26. Kim YS, Kwon OH, Park RH (1999) Wavelet based watermarking method for digital images using the
human visual system. IEE Electronics Letters 35(6):466–468
27. Kumar C, Singh AK, Kumar P (2018) Improved wavelet-based image watermarking through SPIHT.
Journal of Multimedia Tools and Application, Springer:1–14
28. Kutter M, Hartung F (1999) Multimedia watermarking techniques. Proc IEEE 87(7):1079–1107
29. Kutter M, Petitcolas FAP (1999) A fair benchmark for image watermarking systems. Electronic Imaging
‘99, Security and Watermarking of Multimedia Contents, 3657
30. Lee HK, Kim HJ, Kwon SG, Lee JK (2005) ROI medical image watermarking using DWT and bit-plane.
Asia-Pacific Conference on Communications:512–515
31. Li J, Dong C, Han X, et al (2012) DFT based Multiple Watermarks for Medical Image Robust to Common
and Geometrical Attacks. 6th International Conference on New Trends in Information Science and Service
Science and Data Mining (ISSDM), IEEE, p. 472–477
32. Li W, Yang C, Li C, Yang Q (2012) JND Model Study in Image Watermarking”, Advances in Multimedia,
Software Engineering and Computing, 2, (Advances in Intelligence and Soft Computing, 129), p. 535–543
33. Majumdar S, Das TS, Sarkar SK (2011) DWT and SVD based image watermarking scheme using noise
visibility and contrast sensitivity. Int Conf on Recent Trends in Information Technology, IEEE:938–942
34. Makbol NM, Khoo BE (2014) A new robust and secure digital image watermarking scheme based on the
integer wavelet transform and singular value decomposition. Journal of Digital Signal Processing, Elsevier
33:134–147
35. Mishra S, Mahapatra M, Mishra P (2013) A survey on digital watermarking techniques. Int Journal of
Computer Science and Information Technologies 4(3):451–456
36. Mohanty SP. Digital Watermarking: A Tutorial Review. http://www.csee.usf.edu. Accessed 1999
37. Mohanty SP, Bhargava BK (2008) Invisible watermarking based on creation and robust insertion–extraction
of image adaptive watermarks. ACM Trans Multimed Comput Commun Appl 5(2):12:1–12:22
38. Ni R, Ruan Q (2006) Region of interest watermarking based on fractal dimension. Proc Int Conf Pattern
Recogn 3:934–937
39. Niu Y, Kyan M, Ma L, Beghdadi A (2011) A visual saliency modulated just noticeable distortion profile for
image watermarking. European signal processing conference, p. 2039-2043
40. Pareek R, Ghosh PK (2012) Discrete cosine transformation based image watermarking for authentication
and copyright protection. International Journal of Engineering and Advanced Technology (IJEAT) 1(3)
41. Parvez MT, Gutub AA (2011) Vibrant color image steganography using channel differences and secret data
distribution. Kuwait Journal of Science and Engineering 38(1B):127–142
42. Petitcolas FAP, Anderson RJ, Kuhn MG (1999) Information hiding – A survey. Proc of the IEEE 87(7):
1062–1078
43. Prasanna S, Toshanlal M, Ankit S, Vivek G, Amit K (2015) A DWT-SVD based digital watermarking
technique for copyright protection. Int Conf on Electrical, Electronics, Signals, Communication and
Optimization
44. Que D, Zhang L, Lu L, Shi L (2006) A ROI image watermarking algorithm based on lifting wavelet
transform. Proc Int Conf Signal Processing 4. https://doi.org/10.1109/ICOSP.2006.345979
 Author's personal copy
13138 Multimedia Tools and Applications (2020) 79:13125–13138

45. Shaikh S, Deshmukh M (2013) Modulation of watermarking using JND parameter in DCT domain. Int
Journal of Engineering and Advanced Technology 3(2):158–161
46. Sinha Roy S, Basu A, Chattopadhyay A (2019) Perspectives on Digital Image Watermarking, Intelligent
Copyright Protection for Images, 1st edn. CRC, Taylor and Francis, New York
47. Sinha Roy S, Saha S, Basu A (2015) Generic testing architecture for digital watermarking. Proc. FRCCD-
2015, (50-58)
48. Su Q, Chen B (2017) Robust color image watermarking technique in the spatial domain. Springer Journal
on Soft Computing - A Fusion of Foundations, Methodologies and Applications 22(1):91–106
49. Sur A, Sagar SS, Pal R, Mitra P, Mukherjee J (2009) A new image watermarking scheme using saliency
based visual attention model. India Conference, IEEE:1–4
50. The USC-SIPI Image Database. http://sipi.usc.edu/database/database.php?volume=misc
51. Verma M, Yadav P (2013) Capacity and security analysis of watermark image truly imperceptible. Int
Journal of Advanced Research in Computer and Communication Engineering 2(7):2913–2917
52. Wang Q, Yuan Y, Yan P, Li. (2013) Saliency detection by multiple-instance learning. IEEE Transactions on
Cybernetics 43(2):660–672
53. Wong MLD, Lau SIJ, Chong NS, Sim KY (2013) A salient region watermarking scheme for digital
mammogram authentication. Int Journal of Innovation, Management and Technology 4(2):228–232
54. Wu Q, Zhu C, Li JJ, Chang CC, Wang ZH (2016) A magic cube based information hiding scheme of large
payload. Journal of Information Security and Applications 26:1–7
55. Xu H, Wanga J, Kim HJ (2010) Near-optimal solution to pair wise LSB matching via an immune
programming strategy. Inf Sci 180(8):1201–1217
56. Yang CH (2008) Inverted pattern approach to improve image quality of information hiding by LSB
substitution. Pattern Recogn 41(8):2674–2683
57. Zear A, Singh AK, Kumar P (2018) A proposed secure multiple watermarking technique based on DWT,
DCT and SVD for application in medicine. Multimed Tools Appl 77(4):4863–4882

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.