Journal of Ambient Intelligence and Humanized Computing

https://doi.org/10.1007/s12652-020-02135-3

ORIGINAL RESEARCH

A secure image watermarking for tamper detection and localization

Siddharth Bhalerao1 · Irshad Ahmad Ansari1   · Anil Kumar1

Received: 6 September 2019 / Accepted: 15 May 2020

© Springer-Verlag GmbH Germany, part of Springer Nature 2020

Abstract
Data security has become a major concern of present time. Digital image watermarking is seen as a viable solution to ensure
the integrity of image data. Watermarking algorithms can detect the presence of various attacks on images. In this work, a
secure and efficient fragile image watermarking technique is proposed. The proposed watermarking algorithm aims to find
image tampering and its location. The proposed approch is a spatial domain, block-based embedding technique. It introduces
a unique key based embedding using SHA-1 (Secure Hashing Algorithm) hashing. The watermarked images have high PSNR
and SSIM. The algorithm can handle various size of tampering, from very small to large. The proposed technique has been
tested for a wide variety of tampering attacks like copy-paste, copy-move, constant average, and general tampering. It can
efficiently detect tampering in the presence of these attacks.

Keywords  Image watermarking · Tamper detection · Tamper localization · Copy-paste attack · Copy-move forgery ·
Constant average attack

1 Introduction schemes become an important part of multimedia security

It is the age of information and communication technology
that makes data security a prime requirement. The huge
amount of data is flowing through the internet, present in
cloud storage, or stored on hard drives. To curb the illegal
use of multimedia, ownership detection is used. By using
ownership detection techniques, the rightful owner of the
multimedia content could be identifed. Similarly, there is
abundance of softwares, that could easily edit the multime-
dia content. Image and video editors are example of such
softwares. It becomes difficult for the persons with limited
skills to identify any such tampering. In this scenario, such
tampering could affect decisions of individuals or even
organisatios. A tampered medical image or signal may lead
to unthinkable results. Therefore, these tampering detection
* Irshad Ahmad Ansari
irshad@iiitdmj.ac.in
Siddharth Bhalerao
siddharthbhalerao@gmail.com
Anil Kumar
anilkdee@gmail.com
1
Electronics and Communication, PDPM Indian Institute
of Information Technology Design and Manufacturing,
Jabalpur, India
tools.
To ensure data security and integrity, digital cryptogra-
phy, watermarking, and steganography are seen as potential
solutions. In cryptography, different codes are used to secure
the information. In encryption techniques, the impression
of data itself gets changed. The encrypted information can
only be extracted using a valid key. Steganography is another
way to protect the information. Steganography means hid-
den writing. In steganography, a secret message is concealed
within another signal, which could be audio, video, or image
file. Hiding and sharing secret information is one of the
main applications of steganography (Gutub and Al-juaid
2018; Gutub et al. 2019). When secret information is part
of images it is called as digital image steganography (Gutub
et al. 2008; Gutub 2010; Parvez and Gutub 2011). Various
research works (Khan and Gutub 2007; Parvez and Gutub
2008; Gutub and Al-Qahtani 2009; Abu-Marie et al. 2010)
discussed about different image steganography methods.
The watermarking is another way to protect the informa-
tion from being misuse. Watermarking based security comes
with the additional benefit of original data’s shape preserva-
tion. Digital watermarking can be done on any media like
image, audio and video. When watermarking is used to
maintain the integrity of images, it is termed as digital image
watermarking. Digital image watermarking has multiple

13
Vol.:(0123456789)

motives like copyright protection, content authentication,
tamper detection, etc. (Potdar et al. 2005). Apart from it,
watermarking is also used to identify the image tampering
and sometimes to recover the lost information (Wang et al.
2018). Various ways may tamper the images, like noise,
unauthorized access, and illegal activity.
In watermarking, a signal (called as the watermark) is
embedded into a host signal, which is an image in the case
of digital image watermarking. The resultant image is called
a watermarked image. When the requirement is copyright
protection, it is expected that embedded watermark remains
secured even in the presence of attacks on the image. Thus,
watermark can be extracted from the attacked image too and
ownership can be established. Such watermarking is called
robust watermarking (Mohanarathinam et al. 2019; Singh
and Bhatnagar 2019). In fragile watermarking, the main
purpose is identification of image tampering (Di Martino
and Sessa 2019). Thus, when the image gets tampered, the
watermark should also gets damaged, indicating image tam-
pering. Watermarking is a useful tool in the telemedicine
industry, which is used to ensure the integrity of medical
data (Giakoumaki et al. 2006). In the telemedicine industry
patient data is stored electronically in the form of the elec-
tronic patient record (EPR). The major portions of EPR con-
tains medical images like X-rays, CT scans, MRI, etc. (Rao
and Kumari 2011). These images are of sensitive nature and
watermarking is a necessity to identify tampering (Zain and
Clarke 2005).
Watermarking can be performed in the spatial domain
or transform domain (Mousavi et al. 2014). Although there
is no such rule, but transform domain is mostly used, when
robustness is the required characteristic. When it is possible
to extract (from the watermarked image) the host image in
its original (100%) form, then such watermarking is called
reversible watermarking (Arsalan et al. 2012). Watermark-
ing can also be classified as blind and non-blind (Rey and
Dugelay 2002). When watermarking is non-blind, both host
image and watermark are required during data extraction.
A discussion of the earlier developed techniques related to
digital image watermarking is present in the next section.
2 Related work
Different techniques have been reported in the past for digi-
tal image watermarking. Each technique has its specialty.
In spatial domain, the bit substitution is one of the simplest
methods of watermarking (Potdar et  al. 2005). Khamli-
chi (Khamlichi et al. 2006) has proposed a watermarking
method based on LSB substitution. It was a medical image
watermarking technique to preserve the integrity of medical
images. In their technique patient’s confidential data and
13
S. Bhalerao et al.
a message-digest was used as watermark. The digest was
generated from the image itself. Before embedding, encryp-
tion was used to secure the watermark. During extraction,
embedded data was separated from the LSB bit plane. The
digest was again generated from the received image. When
embedded digest and generated digest were found to be
equal, it indicated no tampering was there.
Another spatial domain method for medical image secu-
rity was proposed by (Tan et al. 2011). It was a dual-layer
watermarking scheme with tamper localization ability. Simi-
lar to the above work, this method was also based on cryp-
tography, and it was implemented for DICOM images. The
medical image was firstly fragmented in blocks of 16 × 16
pixels and CRC (cyclic redundancy check) was calculated
for each block. The computed CRC was embedded in the
block itself. The process was repeated during extraction, the
mismatch in CRC indicated tampering. This method was
a reversible scheme with tamper localization ability. How-
ever, embedding CRC of a block in the same block itself
was a drawback of this method. Spatial domain techniques
are mostly used to implement fragile watermarking. To
increase the robustness of watermarking algorithms trans-
form domain techniques were implemented.
Qi and Xin (2015) proposed a transform domain tech-
nique using singular value decomposition (SVD). It was
a semi-fragile technique using the wavelet transform. To
enhance security, the watermark was generated using a
private key. At the time of extraction, the watermark was
regenerated. An error map was created after comparing the
regenerated watermark with the embedded watermark. This
scheme outperformed many of its contemporary schemes,
but its embedding capacity was less.
The security of watermark can be increased using source
coding techniques. A technique using such codes was pro-
posed by Singh et al. (2015). It was a secure and robust
watermarking technique for medical images. It used BCH
codes (Bose, Chaudhuri, and Hocquenghem) to encode the
watermark before embedding it. BCH codes are known for
their error-correcting capabilities. This method was a trans-
form domain approach and DWT (discrete wavelet trans-
form) was used. Watermark was first converted into ASCII
code and then BCH was applied to it, to increase the robust-
ness. Three separate watermarks were embedded in selected
sub-band of a three level wavelet decomposition of medical
image. Their method performed well against a variety of
attacks, providing high imperceptibility and PSNR (peak
signal to noise ratio). However, low payload capacity was
the drawback of this scheme.
A blind watermarking technique for medical images
was proposed in Thakkar and Srivastava (2017). It too was
based on error-correcting codes. It embeds two watermarks
to increase the robustness of the scheme, one of which was
A secure image watermarking for tamper detection and localization
a logo and the other one was text data. Patient data act as
text watermark, which was secured through error-correcting
codes. It was again a DWT based transform domain method.
SVD was applied to the chosen sub-band after decomposi-
tion. They proved that the performance of their method was
better than other methods, but only ROI (region of interest)
was considered for data embedding. This resulted in low
embedding capacity.
Sharma et al. (2017) proposed a DCT (discrete cosine
transform) and DWT based multiple watermarking scheme.
Similar to the previous approach, it used an image and text
watermarks. But in this scheme, both ROI and RONI (region
of non-interest) were utilized for embedding the water-
mark. It used RSA (Rivest, Shamir, and Adleman) encryp-
tion scheme to secure text watermark and the MD5 hashing
algorithm was used for image watermark. To ensure the
robustness of their scheme Hamming error correction codes
were used. Given its complex nature, this method resulted in
high computational complexity. There are many techniques
where SVD is an important ingredient. A method proposed
by Shehab et al. (2018) was also based on SVD. It was a
robust-fragile technique that was able to locate the tamper
and recover the original image. Experiments show that their
technique was effective against various attacks.
Another technique based on IWT was proposed by Sel-
vam et al. (2017). It was a hybrid and blind watermarking
technique implemented using DGT (discrete Gould trans-
form). The host image was first decomposed using IWT
and high-frequency components were selected for water-
mark embedding. The selected sub-band was fragmented in
blocks and DGT was applied on all blocks. Inverse DGT was
performed after embedding watermark bits. In this method
two transform were used. Results showed that such hybrid
methods worked better than single transforms. The downside
of this approach was that it exhibits high distortion with low
payload capacity.
A reversible watermarking scheme with high capacity
was proposed by Parah et al. (2017). In their work, they
have applied pixel-to-block (PTB) conversion method which
guarantees reversibility. Watermark consists of EPR, and
its checksum calculated over each block. By using check-
sum, tamper detection and localization was made possible.
Although reversibility was guaranteed, and high capacity
was achieved, the resultant images were distorted in com-
parison to previous methods.
A real-time tampering detection algorithm was proposed
by Azeroual and Afdel (2017). It was a transform domain
fragile watermarking algorithm for tamper localization.
Watermark was generated by the combination of a logo and
wavelet coefficients. In their method, Faber–Schauder dis-
crete wavelet transform (FSDWT), was used to get the wave-
let coefficients. The watermark was embedded in selected
LSBs of the image. Another method, with a large scale tam-
per detection ability, was proposed by Sarkar et al. (2020). In
their method, two different watermarking schemes were pro-
posed. The first scheme was spatial domain method and the
other one was the transform domain technique. The schemes
were able to recover the tampered image up to 75% and 50%
tampering. In (Di Martino and Sessa 2019), a fuzzy logic
based solution was proposed for image tampering detection.
It was a non-blind technique. Another multipurpose water-
marking technique was proposed in (Tiwari et al. 2017). It
was a vector quantization based approach and two different
watermarks were embedded in the image.
Recently a spatial domain method with high PSNR was
proposed by Gull et al. (2018). It was a tamper detection and
localization technique, implemented for medical images. To
start, the image was first divided into non-overlapping blocks
of 4 × 4. Each block consists of 16 pixels and 16 bits were
embedded in each block. The embedded bits were combina-
tion of watermark and block mean. The method was based
on bit substitution. It was a simple technique that was able
to detect and localize tampering in case of general tamper-
ing. However, their technique failed against specific cases of
copy-paste attack, copy-move attack, and constant average
attack. A detailed security analysis of their method is pre-
sented in the next section. Then after, a solution of the same
is proposed in this work. Different image tampering attacks
and their effects are also discussed in next section.
3 Image tampering attacks
In this section, different image tampering attacks has been
discussed. The performance of Gull et al. (2018) has also
examined for these attacks. The watermarking technique of
(Gull et al. 2018) is explained in Fig. 1 and their perfor-
mance to these attack are present in Fig. 2.
A general tampering attack on an image could be a text or
an object addition. For example, a signature could be added
to the image, or a face could be replaced by another face.
All these attacks are general tampering attacks. The image
present in Fig. 2a is an example of a text addition attack.
Another basic image tampering attack, is crop-attack. In
crop attack, a portion of the image is removed from it. An
example of the crop-attack is present in Fig. 2b. The copy-
paste attack is another well-known image tampering attack.
In copy-paste attack, a part of a different watermarked
image, generated from the same watermarking algorithm
is copied on the other image. A watermarking algorithm is
expected to detect such forgery.
Another attack similar to the copy-paste attack is called
a copy-move attack. This attack is sometimes called a copy-
move forgery. In a copy-move attack, a portion from the
13

Fragmentation
Host Image
4x4 blocks
Fig. 1  Watermark embedding method proposed in Gull et al. (2018)
same watermarked image is pasted on that image itself. In
many methods, image mean is used as the watermark; such
methods are susceptible to constant average attack. In the
constant-average attack, firstly an object is added or removed
from image. Then, the average of the tampered region is
computed and embedded in the tampered region itself.
The watermarking technique proposed by Gull et  al.
(2018) was a spatial domain technique. They were using an
8-bit grayscale image of size 256 × 256 as a host image. The
watermark was also an 8-bit grayscale image of size 64 × 64.
Their embedding scheme is discussed first, followed by the
details of the extraction mechanism. The inherent problems
with this method are also discussed.
During the embedding phase, the host image was initially
divided into non-overlapping blocks of size 4 × 4. As the size
of the host image is 256 × 256; 4096 unique blocks were
created. Given the size of the watermark, the total number
of pixel present in it were 4096, therefore each gray value of
watermark was embedded in one block. Mean (M) of each
block was calculated after replacing the LSB with zeros.
The calculated mean was converted into an 8-bit binary
sequence; ceiling operation was performed, if required. Now,
the block was divided into two halves called the upper half
block (UHB) and the lower half block (LHB). An eight-bit
code was generated by XOR-ing mean (M) and 8 bits of the
watermark. After replacing LSB of every pixel the code was
embedded in LHB itself. The mean (M) was itself embedded
in UHB, by substituting LSB of every pixel of UHB.
In the extraction phase, a similar process was used as
performed during embedding phase. The received water-
marked image was first divided into non-overlapping
blocks of size 4 × 4. The last two bits of every pixel were
replaced with zeros. The block average or mean was calcu-
lated (ext-M). Next, the block was divided into two halves
UHB and LHB. The mean embedded during embedding
(M) was extracted from UHB. The binary code is extracted
from LHB and XOR-ed with the mean (M) to get the water-
mark. When ext-M was found to be same as M, no tam-
pering has been marked. The different values indicated
otherwise. Figure 2a and b shows the performance of this
method against general tampering. The method by Gull
et al. (2018) is prone to the specific case of copy-move
13
S. Bhalerao et al.
Watermark and
Block Mean Image
Mean
Calculation Repackaging
Embedding
Watermark
XOR
attack and copy-paste attack. Similarly, it is susceptible to
the constant-average attack.
In the copy-paste attack, their method fails to identify
tampering. This happens when the part copied from the dif-
ferent image has all 4 × 4 blocks intact i.e. no 4 × 4 block
boundary gets violated. While pasting this block on the tam-
pered image, the same is taken care of, that no 4 × 4 block
boundary gets violated on the image. For example, 32 blocks
arranged in 8 × 4 fashion in one image, will go on 32 blocks
of the same arrangement anywhere on the second image. It
does not need to be at the same location. The output for the
copy-paste attack by Gull et al. (2018) is shown in Fig. 2c.
Similarly, their method fails, when the copy-move attack
is done by preserving the boundaries of each 4 × 4 block.
The reason behind the faulty behavior of this method for
both attacks is the self-authentication of blocks. The aver-
age value of each block is computed and is embedded in the
block itself. Thus, when the tampered region is copied from
the same image (copy-move) or the different watermarked
image (copy-paste). The tampered-blocks self-authenticate
themselves and the method fails to detect tampering. But,
this happens only when attack is done by preserving block
boundaries. The output for the copy-move attack by Gull
et al. (2018) is shown in Fig. 2d.
In the technique proposed by Gull et al. (2018) a block
average can be calculated from LSBs of UHB. The constant-
average attack could be performed by changing the block
and keeping the average value of block equal to the value
extracted from UHB. Alternatively, LSBs of UHB could
be changed according to the tampered block. In both cases,
their method was found unable to detect image tampering. In
Fig. 2e performance of their method against constant average
attack is tested.
To overcome the weakness of above method an improved
method is proposed in this work. In the next section, the
proposed method is discussed.
4 Proposed method
In this section, the proposed method, which is an improved
strategy against the above-mentioned attacks, is discussed.
A secure image watermarking for tamper detection and localization
Fig. 2  Performance of watermarking algorithm of Gull et al. (2018) against different attacks. a Text addition attack. b Crop attack. c Copy-paste
attack. d Copy move attack. e Constant average attack
In the proposed method, a key-based authentication
scheme is implemented for image tampering detection. The
key-based authentication eradicates the problem present
in the self-authentication method. Initially, the host image
was fragmented in non-overlapping blocks of 4 × 4. A 16-bit
block-key was generated for each block. The watermark
for the block was generated using a unique block-key, that
helped in tamper detection of the block. The size of the host
image was 256 × 256; thus, 4096 unique blocks were cre-
ated. The key generated for every block must be same during
embedding and extraction, which is an important condition
for correct extraction. Therefore, block-keys were generated
using the same seed at both ends. The complete process of
the watermark embedding is explained through detailed
steps in the following sub-section.
13

Fig. 3  Example of block embedding
4.1 Embedding steps
(a) The image was first divided into non-overlapping
blocks of size 4 × 4.
(b) A 16-bit block-key was generated for each block. As
the numbers of 16-bit combinations are greater than the
total number of blocks, a unique key gets assigned to
each block.
(c) The LSB of every pixel of the block was replaced with
zero.
Fig. 4  Verification process of blocks
13
S. Bhalerao et al.
(d) The hash value was computed for each stripped block,
using the SHA-1 algorithm. The first 16 bits were
extracted from the hash value. The extracted bits are
named hash-code.
(e) The hash-code was XOR-ed with 16-bit block-key and
a hash-key code was generated.
(f) The generated hash-key code was embedded in the
block itself using LSB substitution.
(g) The block generated after embedding was the water-
marked block.
The process of embedding was performed on each block
as illustrated in Fig. 3. In the Fig. 3, the original host
block is a non-overlapping block of 4 × 4 dimensions with
16 pixels. The SHA-1 hash was calculated after replacing
LSBs of every pixel with zero. The hash-key code was
generated and embedded in the block itself. After embed-
ding on every block, the image was repacked.
4.2 Extraction steps
The extraction process required for tamper detection is
explained through the following steps:
(a) During extraction, the fragmentation on the image was
performed again.
(b) Block-key was generated for each block with the same
seed used during embedding. As the seed was same for
embedding and extraction, the same key was generated
for every block as generated during embedding.
(c) The LSB’s from all 16 pixels of the blocks were
extracted, and replaced with 0. The extracted bits was
hash-key code embedded during the embedding phase.
(d) The hash computation was done for each block after
setting LSBs to zero.
A secure image watermarking for tamper detection and localization

Fragment image in Perform SHA-1

Set LSBs of block Extract 16 bits
Host Image 4x4 non- hash on each
pixels to 0 from block hash
overlapping blocks block

16 bit block-key 16 bit hash

Generate block-
key for every block XOR

Hash-key code

Watermarked Embed hash-key

Repack Image
Image code in block LSB

(a)

Fragment image in Perform SHA-1

Watermarked Replace pixel Extract 16 bits
4x4 non- hash on each
Image LSBs to 0 from block hash
overlapping blocks block

16 bit block-key
16 bit hash
Generate block-
key for every block XOR

Hash-key code

Extract LSBs from

Hash-key code
the block pixels, it
Generated during
is the embedded
extraction
hash-key code

No
Is embedded code equal Tampering
to generated code? Detected

Yes

No Tampering
Detected

(b)

Fig. 5  Block diagram of watermarking algorithm a embedding process, b extraction process

(e) The hash-key code was generated for watermarked
image blocks by XOR-ing block hash codes with block-
keys.
(f) When generated code and embedded code was found to
be same, then it was markred as non- tampered block.
The output (embedded blocks) of Fig. 3 acts as the input
during extraction. Figure 4 shows the validation and veri-
fication of blocks during the extraction process. Initially,
the LSBs are extracted from the block and compared with
regenerated hash-key code. It can be seen from Fig. 4 that
extracted and computed codes are equal, this implies the
absence of tampering.
In Fig. 5, the complete embedding and extraction process
is summarized through block diagrams. Figures 3, 4, and 5
gives the comprehensive idea of proposed work. The pres-
ence of 16-bit block keys makes this implementation resist-
ant to attacks discussed in the previous section. The block
keys are generated by a seed that is common during embed-
ding and extraction. Until and unless this seed is known the
block-keys cannot be calculated. This makes the proposed
improvement to perform well against attacks.

13
 S. Bhalerao et al.

Fig. 6  Performance of proposed method against different attacks. a Copy-move and crop attack, b copy-paste and constant-average attack

Table 1  PSNR and SSIM for common images

Lena Barbara Plane Woman

PSNR SSIM PSNR SSIM PSNR SSIM PSNR SSIM

51.1246 0.9969 51.1366 0.9977 51.1453 0.9963 51.1229 0.9957
Peppers Medical-1 Medical-2 Medical-3

PSNR SSIM PSNR SSIM PSNR SSIM PSNR SSIM

51.1391 0.9968 51.1101 0.9957 51.051 0.9905 51.203 0.9978

5 Results and discussion tamper detection method is designed to work on grayscale

images. All the tests are performed on grayscale images of
In this work, an image tamper detection and localization size 256 × 256. To test the performance, some commonly
solution is proposed. The proposed method can detect image used grayscale and medical images are used. Medical images
tampering in the presence of different types of attacks. The are taken from the same source as used in Gull et al. (2018)

13
A secure image watermarking for tamper detection and localization
Fig. 7  Comparison of proposed method against different attacks. a
General tampering, b copy-paste attack, c copy-move attack, d con-
stant average attack
i.e. the OPENi medical image database. In Fig. 6, the results
of the proposed algorithm, in the presence of different tam-
pering attacks are shown. In Fig. 6a, copy-move and crop
attacks are done on ‘plane’ image. The proposed algorithm
is able to detect and locate tampered regions quite well. In
Fig. 6b, the performance of proposed work in presence of
copy-paste and constant average attacks are shown.
The well-known image quality metrics, like; SSIM (struc-
tural similarity) and PSNR (peak signal to noise ratio) are
used to measure the watermarked image quality. In Table 1,
SSIM and PSNR of watermarked images are presented for
tested host images. A high value of PSNR and SSIM is
obtained, which is evident from Table 1. In Fig. 7, the tam-
per detection and localization ability of the proposed scheme
is compared with the method proposed in Gull et al. (2018).
Both methods are tested against similar tampering attacks.
The proposed scheme is found to be successful in detecting
image tampering and it outperforms the other method.
To find out the effectiveness of the proposed secure and
fragile watermarking scheme, few more parameters are cal-
culated. These parameters are FPR (false positive rate), FNR
(false negative rate), and TDR (tamper detection rate).
Following are the descriptions of the above-mentioned
parameters. The FPR i.e. false-positive rate is the rate at
which the pixels that are not tampered are falsely detected
as tampered pixels:
No. of pixels falsely detected as tampered
Average FPR = × 100
No. of pixeles in tampered region
The FNR i.e. false-negative rate is the rate at which the
pixels that are tampered are falsely detected as not tampered
pixels:
No. of pixels falsely detected as untampered
Average FNR = × 100
No. of pixels in untampered region
The TDR i.e. tamper detection rate is the ratio of pix-
els which are actually tampered pixels to number of pixels
detected as tempered:
No. of detected tampered pixels
Average TDR = × 100
Actual tampered pixels
In Fig. 8, the false positive rate is plotted for different
tampering percentages. The figure shows falsely detected
pixels for random tampering, copy-paste, and copy-move
attacks. The proposed scheme is the block-based method
and even if one pixel from any block is tampered, the com-
plete block will be detected as tampered. Therefore, when
tampering percentage is less, the number of pixels falsely
detected as tampered are more. Initially, the false-positive
rate is high, which quickly descends as tampering percent-
age is increased.
The FPR curves for all the three attacks are almost simi-
lar, this proves that the proposed method is able to detect
tamper equally in the presence of these attacks. The false-
negative rate or FNR is always 0, as the proposed method
can detect 100% of tampered pixels in the presence of all
attacks. Similarly, the tampering detection rate is also 100%
as all tampered pixels are detected as tampered. In Table 2,
the comparison of FPR, FNR, and TDR is presented for dif-
ferent attacks. A comparison of the proposed fragile water-
marking method is done with some previous image tamper
detection techniques in Table 3.
6 Conclusion
The security of multimedia signals has become a prime
concern in this age of digital communication. Therefore,
in present work, a secure and fragile image watermarking
13
 S. Bhalerao et al.

25

Random Tampering
Copy Move
Copy Paste

20

15
False Positive Rate

10

0
0 5 10 15 20 25 30 35 40 45 50

Tampering Percentage

Fig. 8  Average (50 images) false positive rate for different tampering percentage

Table 2  Average (50 images) Tampering per- Random tampering Copy move Copy paste
FPR, FNR and TDR for centage (%)
different tampering attacks FPR FNR TDR FPR FNR TDR FPR FNR TDR

1 20.57609 0 100 23.34487 0 100 22.40706 0 100

5 10.80332 0 100 10.77545 0 100 10.80332 0 100
10 7.526177 0 100 7.53998 0 100 7.544582 0 100
15 5.800525 0 100 6.573646 0 100 6.264667 0 100
20 5.047126 0 100 4.907751 0 100 5.781773 0 100
25 5.149985 0 100 5.021005 0 100 5.389465 0 100
30 3.93685 0 100 4.369657 0 100 4.699268 0 100
35 3.77433 0 100 3.979551 0 100 4.182124 0 100
40 3.342507 0 100 3.434532 0 100 3.246456 0 100
45 3.196016 0 100 3.636846 0 100 3.550095 0 100
50 3.342389 0 100 3.339624 0 100 3.352614 0 100

technique for image tamper detection was proposed. The
proposed technique was a block-based watermarking tech-
nique using the SHA-1 hashing algorithm. Proposed tech-
nique aimed to detect and locate tamper in grayscale images.
The technique was able to detect tampering of all shapes
and sizes. The algorithm was resistant to different image
tampering attacks that were copy-paste, copy-move, and
constant average attacks. The watermarked images showed
high quality, which was evident from their high PSNR and
SSIM. In the future, the algorithm will be extended to incor-
porate features like robustness and self-recovery.

13
A secure image watermarking for tamper detection and localization

Table 3  Comparison of proposed technique with other watermarking algorithms

Method Technique Performance metrics Average Average SSIM Remarks

PSNR
(dB)

Qi and Xin (2015) Spatial domain and singular PSNR 41.39 – Complex algorithm, it was
value decomposition weak towards grometric
(SVD) attacks
Tiwari et al. (2017) Vector quantization PSNR 42 – Output (watermarked image)
PSNR was low
Azeroual and Afdel (2017) Transform domain, Faber– MSE, PSNR, FPR, FNR 51.07 – Number of false positive
Schauder DWT pixels were high
Gull et al. (2018) Spatial domain embedding PSNR, SSIM 51.14 0.9948 It was weak towards copy-
paste, copy-move and
constant average attacks
Sarkar et al. (2020) Spatial domain, quad MSE, PSNR, SSIM 44.5 0.9993 Output (watermarked image)
method PSNR was low
Proposed Spatial domain, hash-based PSNR, SSIM 51.12 0.9959 Detects copy-paste, copy-
embedding move and constant average
attacks

References Khamlichi YI, Zaz Y, Afdel K (2006) Authentication system for medi-
cal watermarked content based image. Wseas Trans Signal Pro-
cess 5:826–830
Abu-Marie W, Gutub A, Abu-Mansour H (2010) Image based steg-
Khan F, Gutub AA-A (2007) Message concealment techniques using
anography using truth table based and determinate array on RGB
image based steganography. In: 4th IEEE GCC Conf Exhib 5
indicator. Int J Signal Image Process 1:9
Mohanarathinam A, Kamalraj S, Prasanna Venkatesan GKD et al
Arsalan M, Malik SA, Khan A (2012) Intelligent reversible water-
(2019) Digital watermarking techniques for image security: a
marking in integer wavelet domain for medical images. J Syst
review. J Amb Intell Humaniz Comput. https​://doi.org/10.1007/
Softw 85:883–894
s1265​2-019-01500​-1
Azeroual A, Afdel K (2017) Real-time image tamper localization
Mousavi SM, Naghsh A, Abu-Bakar SAR (2014) Watermarking
based on fragile watermarking and Faber-Schauder wave-
techniques used in medical images: a survey. J Digit Imaging
let. AEU Int J Electron Commun 79:207–218. https​: //doi.
27:714–729
org/10.1016/j.aeue.2017.06.001
Parah SA, Ahad F, Sheikh JA, Bhat GM (2017) Hiding clinical infor-
Di Martino F, Sessa S (2019) Fragile watermarking tamper
mation in medical images: a new high capacity and reversible data
detection via bilinear fuzzy relation equations. J Amb Intell
hiding technique. J Biomed Inform 66:214–230
Humaniz Comput 10:2041–2061. https​://doi.org/10.1007/s1265​
Parvez MT, Gutub AAA (2008) RGB intensity based variable-bits
2-018-0806-3
image steganography. In: Proc 3rd IEEE Asia-Pacific Serv Com-
Giakoumaki A, Perakis K, Tagaris A, Koutsouris D (2006) Digital
put Conf APSCC 2008, pp 1322–1327. https​://doi.org/10.1109/
watermarking in telemedicine applications-towards enhanced
APSCC​.2008.105
data security and accessibility. In: 2006 International confer-
Parvez MT, Gutub AA-A (2011) Vibrant color image steganography
ence of the IEEE engineering in medicine and biology society.
using channel differences and secret data distribution. Kuwait J
pp 6328–6331
Sci Eng 38:127–142
Gull S, Loan NA, Parah SA et al (2018) An efficient watermarking
Potdar VM, Han S, Chang E (2005) A survey of digital image water-
technique for tamper detection and localization of medical images.
marking techniques. In: INDIN’05. 2005 3rd IEEE international
J Amb Intell Humaniz Comput 1–10
conference on industrial informatics, 2005. pp 709–716
Gutub AA-A (2010) Pixel indicator technique for RGB image steg-
Qi X, Xin X (2015) A singular-value-based semi-fragile watermarking
anography. J Emerg Technol Web Intell 2:56–64
scheme for image content authentication with tamper localization.
Gutub A, Al-Qahtani A (2009) Triple-A: secure RGB image steganog-
J Vis Commun Image Represent 30:312–327
raphy based on randomization. IEEE/ACS Int Conf Comput Syst
Rao NV, Kumari VM (2011) Watermarking in medical imaging for
Appl AICCSA 2009:400–403. https​://doi.org/10.1109/AICCS​
security and authentication. Inf Secur J Glob Perspect 20:148–155
A.2009.50693​56
Rey C, Dugelay J-L (2002) A survey of watermarking algorithms
Gutub A, Al-juaid N (2018) Multi-bits stego-system for hiding text
for image authentication. EURASIP J Adv Signal Process
in multimedia images based on user security priority. J Comput
2002:218932
Hardw Eng 1:1–9. https​://doi.org/10.63019​/jche.v1i2.513
Sarkar D, Palit S, Som S, Dey KN (2020) Large scale image tam-
Gutub A, Ankeer M, Abu-Ghalioun M, et al (2008) Pixel indicator high
per detection and restoration. Multimed Tools Appl. https​://doi.
capacity technique for RGB image based steganography
org/10.1007/s1104​2-020-08669​-0
Gutub A, Al-Juaid N, Khan E (2019) Counting-based secret sharing
Selvam P, Balachandran S, Iyer SP, Jayabal R (2017) Hybrid transform
technique for multimedia applications. Multimed Tools Appl
based reversible watermarking technique for medical images in
78:5591–5619. https​://doi.org/10.1007/s1104​2-017-5293-6
telemedicine applications. Optik 145:655–671

13
 S. Bhalerao et al.

Sharma A, Singh AK, Ghrera SP (2017) Robust and secure multi- Tiwari A, Sharma M, Tamrakar RK (2017) Watermarking based image
ple watermarking for medical images. Wirel Pers Commun authentication and tamper detection algorithm using vector quan-
92:1611–1624 tization approach. AEU Int J Electron Commun 78:114–123. https​
Shehab A, Elhoseny M, Muhammad K et al (2018) Secure and robust ://doi.org/10.1016/j.aeue.2017.05.027
fragile watermarking scheme for medical images. IEEE Access Wang C, Zhang H, Zhou X (2018) A self-recovery fragile image water-
6:10269–10278 marking with variable watermark capacity. Appl Sci 8:548
Singh SP, Bhatnagar G (2019) A robust blind watermarking framework Zain J, Clarke M (2005) Security in telemedicine: issues in watermark-
based on Dn structure. J Amb Intell Humaniz Comput 11:1869– ing medical images. In: Sci Electron Technol Inf Telecommun
1887. https​://doi.org/10.1007/s1265​2-019-01296​-0 Tunis
Singh AK, Kumar B, Dave M, Mohan A (2015) Robust and impercepti-
ble spread-spectrum watermarking for telemedicine applications. Publisher’s Note Springer Nature remains neutral with regard to
Proc Natl Acad Sci India Sect A Phys Sci 85:295–301 jurisdictional claims in published maps and institutional affiliations.
Tan CK, Ng JC, Xu X et al (2011) Security protection of DICOM medi-
cal images using dual-layer reversible watermarking with tamper
detection capability. J Digit Imaging 24:528–540
Thakkar FN, Srivastava VK (2017) A blind medical image watermark-
ing: DWT-SVD based robust and secure approach for telemedi-
cine applications. Multimed Tools Appl 76:3669–3697. https​://
doi.org/10.1007/s1104​2-016-3928-7

13