Computer Communications 152 (2020) 72–80

Contents lists available at ScienceDirect

Computer Communications
journal homepage: www.elsevier.com/locate/comcom

An improved DWT-SVD domain watermarking for medical information

security
Ashima Anand, Amit Kumar Singh ∗
Department of CSE, NIT Patna, Patna, Bihar, India

ARTICLE INFO ABSTRACT

Keywords: Exchange of patient record over network required a technique to guarantee security and privacy for tele-
Watermarking health services. This paper presents an improved watermarking technique capable of providing protection
Medical images of patient data by embedding multi-watermarks in medical cover image using DWT-SVD domain. Prior to
DWT
embedding, Hamming code is applied to text watermark in order to reduce channel noise distortion for
SVD
the sensitive data. After embedding, the watermarked medical image is encrypted then compressed. Out of
ECC
Chaotic
two encryption method and three compression scheme tested, combination of Chaotic-LZW shows the best
LZW performance. However, HyperChaotic-LZW combination is more robust against Gaussian, JPEG compression,
speckle noise and histogram equalization attacks. We illustrate the good results in terms of objective and
subjective evaluation, and verify its robustness for various attacks while maintaining imperceptibility, security
and compression ratio. Experimental results demonstrate that the suggested technique archives high robustness
against attacks in comparison to the other scheme for medical images.

1. Introduction
Some potential applications of digital watermarking are depicted in
Fig. 1 [6,8,9].
Security and privacy of medical information are mostly presented Currently, encryption and watermarking approaches are used to
as major issue in teleHealth services [1]. The teleHealth services uses provide authenticity and confidentiality of medical record [10–13]. The
advanced cost effective information and communication technology watermarked media being transmitted might get some unwanted chan-
(ICT) tools to manage, store and transmit medical record for vari- nel noises or get attacked upon. This makes the technique less robust
ous purpose [2]. These services are convenient and potentially useful resulting in receiving distorted media. So the concept of encryption
in medical domain but also bring a privacy leakage, identity theft is combined with watermarking to make the technique anti-attack in
and risk of data tampering and authenticity at the same time [3]. nature [14,15]. Encryption deals with the secure transmission of the
Further, digital imaging and communications in medicine (DICOM) data over an insecure transmission media [16].
is used as standard to exchange the electronic patient report over In order to this, many researchers used combined (Encryption-
open network [4]. A header file is attached with DICOM image which Watermarking) approach to address the security issues of teleHealth ap-
contains significant patient related information. However, this header plications [4,10–13]. However, few studied of watermarking-
consumes extra bandwidth and may be altered, removed or attacked encryption-compression concepts for the applications [14,17]. In this
during the transmission. Furthermore, medical related identity theft is paper, we developed an improved dual watermarking approach that
booming problem as demonstrated by various surveys [3,5]. According uses combination of DWT-SVD, Encryption-compression and hamming
to a survey conducted by Experian Information Solutions in 2017, code to provide security for medical information exchanged via unse-
medical thefts constitute more than 27% of the total identity thefts [5]. cured networks. The approach uses multi-watermarks embedded into
The stolen information affects the individual life to a great extent. SVD of cover image in DWT domain. The DWT has great spatio-
Therefore, there is a need for strong mechanisms to be proposed which frequency properties which help in recognizing the area where the
resolves the security and privacy issue of medical information for watermark can be embedded with high imperceptibility. Also, embed-
teleHealth services. Recently, digital watermarking techniques have ding in the singular matrix of SVD does not have noticeable change
attracted remarkable interest for protection of medical data [3,6]. The in the visual quality of the cover image. Thus, using a hybrid of DWT
technique refers the embedding of confidential information in a me- and SVD helps in achieving higher robustness and imperceptibility [4].
dia/entity for determining any ownership conflicts that can arise [3,7]. An encoded text watermark via hamming code is also embedded in the

∗ Corresponding author.
E-mail addresses: [email protected] (A. Anand), [email protected] (A.K. Singh).

https://doi.org/10.1016/j.comcom.2020.01.038
Received 25 November 2019; Received in revised form 9 January 2020; Accepted 18 January 2020
Available online 22 January 2020
0140-3664/© 2020 Elsevier B.V. All rights reserved.
A. Anand and A.K. Singh Computer Communications 152 (2020) 72–80

visual feature vector of the host image which is extracted using SWT-
DCT. Also, to improve the authenticity, the watermark is encrypted
using chaotic map. This watermarking technique is robust against
conventional and geometric attacks. It was experimentally proved that
on embedding more data, the pixel value of the host image does
not change and maintains low complexity. In [23], three watermarks
are embedded in the RONI part of transformed DWT cover. RNOI is
obtained by sub-band decomposition of the host image. Prior hiding,
error correcting code (ECC) is used to provide better robustness of the
hidden data. A biometric based watermarking technique was proposed
in [13] for protection of medical data. In this method, minutia feature
of the fingerprint and encrypted EPR report is considered as watermark.
These watermarks are embedded in the ROI part of the cover. The
resulting watermark image is further compressed by arithmetic coding
scheme to generate the final watermark image. The final watermark
data and a pseudo random number id are placed inside the cover image.
Different results showed the proficiency of the scheme for medical data
authentication and reliability of the system. Soualmi et al. developed a
Fig. 1. Novel applications of digital watermarking.
blind watermarking for securing the medical data [24]. It uses a fusion
of DWT-Schur decomposition for hiding the medical data. Also, to
improve the robustness, each non-overlapping block of cover image is
higher level DWT to provide more security. Additionally, the water- scrambled. This technique offers better robustness and imperceptibility
marked image is first encrypted then compressed prior transmission to when compared with similar proposed techniques [25–29].
achieve extra security and reduce the bandwidth demand. Therefore, DWT-SVD based robust and secure watermarking technique is pro-
the major advantages the of our technique are (1) Robustness and posed by Singh et al. for medical images [30]. The image watermark
imperceptibility is increased by using hybrid transform (DWT-SVD) is hidden in the RNOI region of the cover image, offering better
at low distortion, (2) Providing extra level of security by embedding imperceptibility and hence reducing the distortion in the medical cover
multiple watermarks, (3) Encryption of the marked image is providing image.
confidentiality at the same time, and (4) Compression of encrypted data A robust watermarking approach is proposed in [4]. Two water-
shown the good compression performance. marks are placed inside the cover using a hybrid form of DWT-SVD.
Rest of the paper is organized as follows. Recent work related to Prior the embedding procedure, the image watermark was sub-divided
teleHealth applications is introduced in Section 2. Section 3 describes into two parts and text watermark is encoded using error correcting
the more important concepts related to our work. The embedding and codes to increase the robustness. Four types of ECC were studied to
recovery part of our method is elaborated in Section 4. Section 5 determine the performance of the technique. The sub-part of the image
presents the performance analysis of the work. Finally, conclusions are watermark data and encoded text is imperceptibly placed inside the
drawn in Section 6. transformed cover media.

3. Preliminary
2. Literature review

3.1. Discrete wavelet transforms (DWT)

Some of the related work is proposed in literature survey are discuss
below.
The DWT is used to decompose the image into non-overlapping
Least Significant Bit (LSB) based data hiding scheme for colored
sub-bands, namely, approximation (LL), horizontal (LH), vertical (HL)
medical images is introduced by Boonyapalanant et al. in [18] in which
and diagonal (HH). We can further decompose the ‘LL’/or other sub-
QR code containing the entire patient’s data is used as watermark. The band to obtain wavelet coefficients at multiple level (see Fig. 2).
embedding of the watermark is distributed over the whole image and Multi-resolution property of DWT helps in easily locating the areas
does not affect the image size. It is experimentally proved that this in the cover image where the watermark should be embedded [3].
technique is robust against rotational attacks. A blind, robust and zero Above mentioned properties makes DWT suitable for medical image
watermarking technique based on Fast Discrete Curvelet Transform watermarking [4,11,22,23]. It is important to notice that ‘LL’ sub-band
(FDCT) and perceptual hash is proposed in [10]. Its security factor is is more sensitive to human eye, in the sense that any operation in this
improved by applying logistic chaotic encryption on the watermark. sub-band may be noticeable to eye. However, this sub-band is used to
This encrypted data is placed in the low-frequency coefficients of achieve good robustness [31]. Further, it is proved that ‘HH’ sub-band
the first level FDCT. In terms of imperceptibility and robustness, this is better choice for good imperceptibility. However, it has been recently
technique experimentally proven to be superior to techniques proposed shown that this sub-band offered poor robustness. Basically, 2D wavelet
in [19,20]. A blind watermarking technique is proposed by Araghi transform is mathematically represented as [32],
et al. in [11] using DWT and second level SVD. To provide better
𝐿𝐿 = 𝛼 (𝑥, 𝑦) = 𝛼 (𝑥) 𝛼 (𝑦) (1)
protection to the watermarked image, encryption is applied to it. It
provides biometric based source protection. Watermarked image is 𝐻
𝐿𝐻 = 𝛽 (𝑥, 𝑦) = 𝛽 (𝑥) 𝛼(𝑦) (2)
visually encrypted with the fingerprint image using the IWT. It helps in 𝑉
𝐻𝐿 = 𝛽 (𝑥, 𝑦) = 𝛼 (𝑥) 𝛽 (𝑦) (3)
preventing both the intentional and unintentional attacks. It provides
𝐷
better security and authentication in comparison to [21]. A watermark- 𝐻𝐻 = 𝛽 (𝑥, ) = 𝛽 (𝑥) 𝛽(𝑦) (4)
ing technique based on RONI is proposed in [22] for improving the
Here, H, V and D represent the direction of decomposition of the
authenticity of the medical data. GUI based approach is used to sepa-
wavelet. Scaling function (𝛼) and wavelet functions (𝛽) are given as:
rate the ROI from the rest of the image. Multiple character watermarks 𝑖
are placed inside the cover using fourth level lifting based wavelet 𝛼𝑖,𝑚,𝑛 (𝑥, 𝑦) = 2 2 .𝛼(2𝑖 .𝑥 − 𝑚, 2𝑖 .𝑦 − 𝑛) (5)
transform. It makes the technique more robust and secure. A zero wa- 𝑗 𝑖
𝛽𝑖,𝑚,𝑛 (𝑥, 𝑦) = 2 .𝛽 𝑗 (2𝑖 .𝑥 − 𝑚, 2𝑖 .𝑦 − 𝑛)
2 (6)
termarking technique is proposed in [12] based on discrete stationary
wavelet transform (SWT) and DCT. The watermark is embedded in the where, j denotes the direction i.e. H, V or D.

73
A. Anand and A.K. Singh Computer Communications 152 (2020) 72–80

Fig. 2. Two level DWT decomposition of an image.

Fig. 3. Representation of (7, 4) hamming code.

Table 1
Properties of U, V and S matrices [35].
Matrix Size Type of matrix Features
U m × m Unitary, ∙ Represents set of output vector direction
Orthogonal for input matrix
Unitary, ∙ Conjugate transpose of V.
VT n × n
Orthogonal ∙ Represents set of input vector direction
for input matrix
Diagonal, ∙ Non-negative numbers in the diagonal.
S m × n
singular ∙ Represents singular value of input image
in descending order i.e. 𝜎i > 𝜎i+1

Table 2
Relationship between parity bits and data bits.
Parity bit Relationship with data bits Fig. 4. Encryption then compression technique [37].
p1 d1 ⊕d2 ⊕d4
p2 d1 ⊕d3 ⊕d4
p3 d2 ⊕d3 ⊕d4
calculated using odd or even parity of the associated bits. These bits
are then used to identify one or two bits error [36] by ensuring that
each parity bit and their corresponding data bits together achieve even
3.2. Singular value decomposition (SVD) parity. Fig. 3 shows the representation of (7, 4) hamming code.

SVD reduces an image of any size into smaller invertible and square 3.4. Compression of encrypted data
matrices. It has good energy compaction properties. SVD of a matrix X
of size m × n can be stated as [33]: Encryption Then Compression (ETC) technique helps for transfer-
ring the image more securely over the unprotected network. Sender
protects the privacy of the data by encrypting it before handing over
X = U S VT (7)
to the network provider. He also compresses the encrypted data without
Here, U and V are orthogonal matrices and S is a diagonal matrix knowing the encryption key, resulting in reduced network traffic. The
representing the singular values of X in descending order. SVD can be receiver will obtain the original data by performing joint decompres-
applied to all sizes of image including square and rectangular images, sion and decryption on the received data. This method is suitable for
making it one of the widely used forms of transform in watermarking two work plans; (a) when the storage service provider wants to reduce
technique. Also, slight alterations in the intensity values of the singular the memory usage without any knowledge of the encryption key, and
values do not make significant variations in the visual quality of the (b) when there is a strong need to secure transmission through an
image [34]. The properties of U, V and S matrices are described in untrusted channel provider [14,15]. This technique is represented in
Table 1. Fig. 4.
In our study, two different encryption scheme (Chaotic and Hy-
3.3. Error correcting code (ECC) perChaotic) and three lossless compression techniques (LZW, Huffman,
and a hybrid of Huffman and LZW) are considered. Hence, some small
During the transmission of the data over the communication chan- background is given below.
nel, noises may introduce. Hence, small noisy data may cause big Chaotic system [38], developed in 1989, is a non-linear system which
problems in some sensitive applications [36]. Note that different error is invertible and dynamic in nature. The chaotic sequence can be
correcting methods have been introduced to correct the errors such modeled using non-linear logistic maps, defined as
[ ]
as repetition codes, Hamming codes, BCH codes and Reed–Solomon Ci+1 = 𝜎 Ci 1 − Ci (8)
codes [36,37].
where, Ci ∈ (0, 1) ; 𝜎 ∈ (0, 4)
In our method, hamming code (7, 4) is used where four data bits (d1 ,
d2 , d3 and d4 ) are represented using seven bits. Rest three bits are the Here, 𝐶𝑖+1 and 𝜎 represents the chaotic sequence and bifurcation
parity bits (p1 , p2 and p3 ). Relationships between these bits are depicted factor respectively. This sequence will be obtained by iterating for
in Table 2. Each parity bit is generated by applying exclusive or (XOR), a specified amount of number. XOR operation of the binary form
denoted by ‘‘⊕’’, on three data bits. Value of each parity bit added is of chaotic sequence and the image produced the encrypted image.

74
A. Anand and A.K. Singh Computer Communications 152 (2020) 72–80

4. Proposed method

The proposed method consists of mainly three parts i.e. embedding,

transmission and extraction of the watermarks. Before the embedding
process, the text watermark is encoded using Hamming code and the
image watermark is divided into two sub images. Cover image is
then decomposed at second level DWT. Singular component of in-
termediate frequency sub-bands i.e. ‘HL’ and ‘LH’, of the first level
DWT coefficients are embedded with the two sub-parts of the medical
watermark images. Further, the encoded text watermark is embedded
in the ‘HH’ sub-band of the second level DWT coefficients. Inverse SVD
and inverse DWT is finally applied to obtain the marked image. After
embedding, the marked medical image is encrypted then compressed.
This encrypted then compressed marked image is then transferred
over communication channel. The discussed process including the dual
watermark embedding, transferring and extraction is illustrated using
Fig. 6. Mentioned below is the detailed procedure of embedding and
transmission algorithm. The extraction procedure is the exact reverse
of the mentioned procedure.
Embedding algorithm
Step 1: Consider a MRI image of size 512 × 512 as cover image
(cover_img) and apply second level DWT on it.
[ ]
𝐷𝑊 𝑇 (𝑐𝑜𝑣𝑒𝑟_𝑖𝑚𝑔, ‘𝐻𝑎𝑎𝑟’) = 𝐿𝐿1 , 𝐻𝐿1 , 𝐿𝐻1 , 𝐻𝐻1 (10)
[ ]
𝐷𝑊 𝑇 (𝐿𝐿1 , ‘𝐻𝑎𝑎𝑟’) = 𝐿𝐿2 , 𝐻𝐿2 , 𝐿𝐻2 , 𝐻𝐻2 (11)

Step 2: Perform SVD on HL1 (horizontal), LH1 (vertical) sub-bands.

( ) [ ]
𝑆𝑉 𝐷 𝐻𝐿1 = 𝑈𝐻𝐿1 , 𝑆𝐻𝐿1 , 𝑉𝐻𝐿1 (12)
( ) [ ]
𝑆𝑉 𝐷 𝐿𝐻1 = 𝑈𝐿𝐻1 , 𝑆𝐿𝐻1 , 𝑉𝐿𝐻1 (13)
Fig. 5. Flow chart of LZW compression technique.
Step 3: Read the image watermark (watermark_img) of size 256 × 256
and divide this image into two equal halves.

The non-periodic, uncorrelated and pseudorandom features make the 𝑊 𝑎𝑡𝑒𝑟𝑚𝑎𝑟𝑘_𝑖𝑚𝑔 = 𝑊1 + 𝑊2 (14)
chaotic system appropriate for encryption. Since chaotic system only
Step 4: Embed the two parts of the watermark image into singular
diffuses in single direction, HyperChaotic system was introduced. Hy-
matrices (SHL1 , SLH1 ) obtained in Step 2 using a gain factor (𝛼).
perChaotic system [39,40] diffuses in at least two directions, which
results in more than one positive Lyapunov exponent. This property 𝑊 𝑆𝐻𝐿1 = 𝑆𝐻𝐿1 + 𝛼 ∗ W1 (15)
makes it more robust against the attacks. Higher number of posi- 𝑊 𝑆𝐿𝐻1 = 𝑆𝐿𝐻1 + 𝛼 ∗ W2 (16)
tive Lyapunov exponent has high randomness, which makes the sys-
tem more complex and anti-attack. The mathematical definition of Step 5: Apply inverse SVD to obtain the watermarked DWT sub-bands.
HyperChaotic system is defined as, ( )𝑇
𝑊 𝐻𝐿1 = 𝑈𝐻𝐿1 ∗ 𝑊 𝑆𝐻𝐿1 ∗ 𝑉𝐻𝐿1 (17)
( )𝑇
⎧ 𝑊 𝐿𝐻1 = 𝑈𝐿𝐻1 ∗ 𝑊 𝑆𝐿𝐻1 ∗ 𝑉𝐿𝐻1 (18)
⎪ 𝑎̇ = 𝑝(𝑏 − 𝑎)
⎪𝑏̇ = (𝑟 − 𝑝) 𝑎 − 𝑎𝑐 + 𝑟𝑏 + 𝑑 Step 6: Read the text watermark and convert it into binary form.
⎨ (9)
⎪ 𝑐̇ = 𝑎𝑏 − 𝑞𝑐
⎪ 𝑑̇ = −𝑖𝑎 + 𝑗𝑏 𝑇𝑏𝑖𝑛 = 𝑏𝑖𝑛𝑎𝑟𝑦(‘Ashima NITP01’) (19)
⎩
Step 7: Apply Hamming error correcting code on Tbin to obtain the
where, a (0) = b(0) = c(0) = d(0) = 1, p = 35, q = 3, r = 28, i = 2, j =
encoded message.
0.5.
Lempel–Ziv–Welch (LZW) [41] is a lossless dictionary-based com- 𝑇𝑒𝑛𝑐𝑜𝑑𝑒𝑑 = 𝐻𝑎𝑚𝑚𝑖𝑛𝑔𝐶𝑜𝑑𝑒(𝑇𝑏𝑖𝑛 ) (20)
pression technique used for both images and text data. Unlike some
Step 8: Replace (0,1) with (−1,1) in the encoded bits to obtain Tf inal
compression methods, the code length is fixed in LZW compression and embed it into HH2 obtained in Step 1.
and additional information or probability density is not required to
decompress the media. The LZW encoding algorithm is depicted in 𝑊 𝐻𝐻2 = 𝑡𝑒𝑥𝑡_𝑒𝑚𝑏𝑒𝑑(𝐻𝐻2 , 𝑇𝑓 𝑖𝑛𝑎𝑙 , 𝐿, 𝛼1 ) (21)
Fig. 5 [42]. where, L is the length of Tf inal and 𝛼1 is the gain factor used to embed
Huffman compression [43] technique is based on the frequency of the text.
occurrence of a symbol in original file. Higher the frequency, shorter Step 9: 𝑊 𝑎𝑡𝑒𝑟𝑚𝑎𝑟𝑘𝑒𝑑_𝑖𝑚𝑔 (marked image) is obtained via inverse-DWT.
is the length of bit string representing that symbol. Huffman-tree
represents the relationship between the original symbols and their 𝑊 𝐿𝐿1 = 𝐼𝐷𝑊 𝑇 (𝐿𝐿2 , 𝐻𝐿2 , 𝐿𝐻2 , 𝑊 𝐻𝐻2 ) (22)
corresponding bit strings. Both compression and decompression pro- 𝑊 𝑎𝑡𝑒𝑟𝑚𝑎𝑟𝑘𝑒𝑑_𝑖𝑚𝑔 = 𝐼𝐷𝑊 𝑇 (𝑊 𝐿𝐿1 , 𝑊 𝐻𝐿1 , 𝑊 𝐿𝐻1 , 𝐻𝐻1 ) (23)
cess is carried out using this Huffman tree. This tree is formed by
Step 10: Encrypt the watermarked image using either Chaotic or Hy-
sorting the symbols in descending order of their frequencies and then
perChaotic encryption algorithm.
merging the frequencies together until all frequencies are merged. For
decompression, reverse compression process is performed. 𝐸𝑛𝑐𝑖𝑚𝑔 = 𝐸𝑛𝑐𝑟𝑦𝑝𝑡𝑖𝑜𝑛(𝑊 𝑎𝑡𝑒𝑟𝑚𝑎𝑟𝑘𝑒𝑑_𝑖𝑚𝑔) (24)

75
A. Anand and A.K. Singh Computer Communications 152 (2020) 72–80

Fig. 6. Proposed watermarking method.

76
A. Anand and A.K. Singh Computer Communications 152 (2020) 72–80

Fig. 7. Sample of our images and text data as a. cover image, b. watermark image, c. watermarked image, d. text watermark.

Table 3
Various metrics used for objective measures.
Standard metric Formula Description
(255)2
Peak Signal to Noise 𝑃 𝑆𝑁𝑅 = 10 𝑙𝑜𝑔 where, Mean Square Error (MSE) -Determines the similarity between original and
𝑀𝑆𝐸
Ratio (PSNR) [8] 1 ∑𝑋 ∑𝑌 2 watermarked image.
is, 𝑀𝑆𝐸 = 𝑋×𝑌 𝑖=1 𝑗=1 (𝐼𝑖𝑗 − 𝑊𝑖𝑗 )
Iij = Pixel of original image of size X × Y -Higher the PSNR value, greater the similarity between
𝑊ij = Pixel of watermarked image of size X × Y. the two images.
-Ideally, the value of PSNR should be more than 28 dB.
∑𝑋 ∑𝑌
𝐼=1 𝑗=1 (𝑊𝑜𝑟𝑔𝑖𝑗 ×𝑊𝑟𝑒𝑐𝑖𝑗 )
Normalized Coefficient 𝑁𝐶 = ∑𝑋 ∑𝑌 ( 2 ) -Calculates similarity between original and extracted
𝐼=1 𝑗=1 𝑊𝑜𝑟𝑔_𝑖𝑗
(NC) [8] watermark.
Worigij = Pixel of original watermark of size X × Y
-Its value lies between 0 and 1
𝑊recij = Pixel of recovered watermark of size X × Y.
-Acceptable above 0.7 value.
-Determines the robustness.
(𝑁𝑜.𝑜𝑓 𝑖𝑛𝑐𝑜𝑟𝑟𝑒𝑐𝑡𝑙𝑦𝑑𝑒𝑐𝑜𝑑𝑒𝑑𝑏𝑖𝑡𝑠)
Bit Error Rate (BER) 𝐵𝐸𝑅 = -Ideally it should be zero.
(𝑇 𝑜𝑡𝑎𝑙𝑛𝑢𝑚𝑏𝑒𝑟𝑜𝑓 𝑏𝑖𝑡𝑠)
[4]
{
0, 𝑖𝑓 𝐶 1 (𝑖, 𝑗) = 𝐶 2 (𝑖, 𝑗)
𝐷 (𝑖, 𝑗) =
0, 𝑖𝑓 𝐶 1 (𝑖, 𝑗) ≠ 𝐶 2 (𝑖, 𝑗)
∑ 𝐷 (𝑖, 𝑗)
1 2
Number of Changing 𝑁𝑃 𝐶𝑅 ∶ N(𝐶 , 𝐶 ) = -Used evaluate the efficiency of the image encryption
𝑖,𝑗
𝑇
Pixel Rate (NPCR) and algorithm against different attacks.
Unified Averaged ∑ |𝐶 1 (𝑖, 𝑗) − 𝐶 2 (𝑖, 𝑗) | -Higher the NPCR/UACI value, higher the resistance
𝑈 𝐴𝐶𝐼 ∶ U(𝐶 1 , 𝐶 2 ) =
Changed Intensity 𝑖,𝑗
𝐹 .𝑇 provided against the attacks.
(UACI) [44] where,
C1 ,C2 = Ciphertext images before and after one pixel
change,
F = Largest supported pixel value,
T = Total number of pixels in ciphertext image.
𝑆𝑆𝐼𝑀 = 𝑓 (𝑎 (𝑚, 𝑛) , 𝑏 (𝑚, 𝑛) , 𝑐(𝑚, 𝑛))
2𝜇𝑚 𝜇𝑛 + 𝐶1
𝑎 (𝑚, 𝑛) =
𝜇𝑚2 + 𝜇𝑛2 + 𝐶1
Structural Similarity 2𝜎𝑚 𝜎𝑛 + 𝐶2 -Used to find how similar are the watermarked and
Index (SSIM) [8] 𝑏 (𝑚, 𝑛) = original image with each other.
𝜎𝑚2 + 𝜎𝑛2 + 𝐶2
-Value ranges from −1 to +1. Ideally the value should be
𝜎𝑚𝑛 + 𝐶3
𝑐 (𝑚, 𝑛) = + 1.
𝜎𝑚 𝜎𝑛 + 𝐶 3
where,
a(m,n)= luminance comparison function,
b(m,n) = contrast comparison function,
c(m,n)= structure comparison function
(𝑆𝑖𝑧𝑒𝑎𝑓 𝑡𝑒𝑟𝑐𝑜𝑚𝑝𝑟𝑒𝑠𝑠𝑖𝑜𝑛)
Compression Ratio [45] 𝐶𝑅 = -Ratio of size of data after compression and size of data
(𝑆𝑖𝑧𝑒𝑏𝑒𝑓 𝑜𝑟𝑒𝑐𝑜𝑚𝑝𝑟𝑒𝑠𝑠𝑖𝑜𝑛)
before compression.
-Value to be less than 1 and more than 0.
-Value approaching 0 is better.

77
A. Anand and A.K. Singh Computer Communications 152 (2020) 72–80

Table 4
Performance analysis at varying gain.
Technique used
Gain Factor PSNR (in dB) NC BER NPCR UACI SSIM CR
Encoding Compression
0.05 36.1007 0.9851 0 0.9954 0.3865 0.996153 0.9976
Huffman
0.10 35.8572 0.9911 0 0.9954 0.3865 0.995939 0.9976

Chaotic 0.05 36.1007 0.9851 0 0.9954 0.3865 0.996153 0.9089

LZW
0.10 35.8572 0.9911 0 0.9954 0.3865 0.995939 0.9084
0.05 36.1007 0.9851 0 0.9954 0.3865 0.996153 0.9407
Hybrid
0.10 35.8572 0.9911 0 0.9954 0.3865 0.995939 0.9413
0.05 34.2648 0.9721 0 0.9961 0.3964 0.996098 1
Huffman
0.10 33.8755 0.9757 0 0.9961 0.3965 0.995969 1

Hyper chaotic 0.05 34.2648 0.9721 0 0.9961 0.3964 0.996098 0.9466

LZW
0.10 33.8755 0.9757 0 0.9961 0.3965 0.995969 0.9466
0.05 34.2648 0.9721 0 0.9961 0.3964 0.996098 0.9466
Hybrid
0.10 33.8755 0.9757 0 0.9961 0.3965 0.995969 0.9466

Table 5
Performance analysis for various medical and non-medical images.
Cover Image ETC technique PSNR (in dB) NC BER NPCR UACI SSIM CR
Chaotic-LZW 38.3532 0.9994 0 0.9961 0.3922 0.997860 0.9413
MRI
Hyper chaotic-LZW 38.0974 0.9933 0 0.9959 0.4004 0.997209 0.9475
Chaotic-LZW 31.1201 0.9860 0 0.9960 0.3505 0.994665 0.9408
Kidney stones
Hyper chaotic-LZW 28.9437 0.9643 0 0.9960 0.3528 0.993634 0.9470
Chaotic-LZW 33.1555 0.9873 0 0.9954 0.3867 0.996885 0.9221
Colon MRI
Hyper chaotic-LZW 32.1846 0.9695 0 0.9961 0.3882 0.995945 0.9468
Chaotic-LZW 36.1058 0.9835 0 0.9953 0.3865 0.996157 0.9085
Head CT scan
Hyper chaotic-LZW 34.2674 0.9668 0 0.9961 0.3964 0.996074 0.9466
Chaotic-LZW 29.7203 0.9991 0 0.9960 0.2957 0.988501 0.9462
Barbara
Hyper chaotic-LZW 29.2983 0.9937 0 0.9961 0.2991 0.985800 0.9473
Chaotic-LZW 32.3953 0.9982 0 0.9961 0.3059 0.995167 0.9455
Camera man
Hyper chaotic-LZW 31.4553 0.9939 0 0.9959 0.3112 0.994635 0.9458
Chaotic-LZW 44.1232 0.9997 0 0.9948 0.2353 0.993538 0.9249
Cell
Hyper chaotic-LZW 44.1944 0.9942 0 0.9959 0.2556 0.991374 0.9463
Chaotic-LZW 36.3285 0.9999 0 0.9957 0.2738 0.995821 0.9457
Rice
Hyper chaotic-LZW 36.3419 0.9716 0 0.9962 0.2826 0.994264 0.9467

Step 11: Compress Encimg using Huffman compression or LZW com- The highest PSNR values as 44.1944 dB is observed for Cell image using
pression or hybrid of both compressing techniques. Hyperchaotic-LZW technique. BER is 0 for all images with every ETC
technique. Best value of CR, NC, NPCR, UACI and SSIM are 0.9085,
𝐶𝑜𝑚𝑝𝑟𝑒𝑠𝑠𝑒𝑑_𝑑𝑎𝑡𝑎 = 𝑐𝑜𝑚𝑝𝑟𝑒𝑠𝑠𝑖𝑜𝑛(𝐸𝑛𝑐𝑖𝑚𝑔 ) (25)
0.9999, 0.9961, 0.3865 and 0.996153 respectively.
Further, we have tested our method on hundred dissimilar medical
5. Result and analysis
cover images [46]. Table 6 shows the average value of performance
parameters obtained during the validation. Using chaotic-LZW tech-
We conducted experiments on hundred different medical cover
nique, higher value of average PSNR and SSIM is obtained. However,
images (size: 512 × 512) [46], watermark data: image of size 256 × 256
average values for NPCR, UACI and NC are better for Hyperchaotic-
and text of size up to 12 characters, using MATLAB R2017a. Addition-
LZW technique. BER is 0 for both the cases. The subjective measure is
ally, ‘‘Ashima NIT01’’ is taken as text watermark. Fig. 7 presents the
formulated by combining the opinions of significant amount of people
some samples of our images and text data. The objective and subjective
method [3] are used to determine the performance of our approach. and is tabulated in Table 7. With small value gain factor (𝛼 = 0.001),
The detail description of all considered standard metric (objective supreme quality of watermarked image is obtained. As gain factor in-
measures) is provided in Table 3. creases, the quality of watermarked image degraded, making it difficult
The proposed method is simulated on different values of gain factor for diagnosis.
(𝛼) as 0.05 and 0.10. The results are shown in Table 4. In this Table, two Three different size of text watermark along with cover image
different encoding techniques (chaotic and hyper chaotic encoding) are shown in Fig. 7 are considered to evaluate the standard metrics, as
used with three compression techniques including Huffman, LZW and depicted in Table 8. It can be noticed the highest PSNR value is 36.1058
a hybrid of Huffman and LZW techniques. Without any noise attack, dB with BER = 0. Further, value of NC and SSIM are also very close to
highest PSNR = 36.1007 dB is obtained using chaotic encryption with 1. Hence, we have robustly recovered all hidden data from the cover.
𝛼 = 0.05 with highest NC = 0.9911 at 𝛼 = 0.10 and BER=0. It is observed Tables 9 and 10 present the robustness analysis of the proposed work
that on increasing the value of gain factor the robustness increases using Chaotic-LZW and Hyperchaotic-LZW, respectively. It is observed
but the image quality decreases. NPCR and UACI are approximately that the technique is robust against salt and pepper noise, Gaussian
reaching the acceptable values for all gain factors. Table 5 provides noise, JPEG compression, speckle noise, median filter and histogram
the performance for eight dissimilar cover images at gain value = 0.05. attacks. Value of BER is obtained as ‘0’ expect for rotation and cropping

78
A. Anand and A.K. Singh Computer Communications 152 (2020) 72–80

Table 6 Table 9
Average values of the performance parameters when tested on hundred medical Performance of NC and BER value for different attacks using chaotic-LZW technique.
images. Noise density Attack NC (image) BER (text)
Performance Parameter Average value for each measure
0.0001 0.9879 0
Chaotic_ LZW Hyper-Chaotic_ LZW 0.0005 Salt and pepper noise 0.9693 0
0.001 0.9251 0
PSNR (dB) 34.0455 32.6229
NPCR 0.9953 0.9961 0.0001 0.9803 0
Gaussian noise
UACI 0.3859 0.3914 0.0005 0.8761 0
SSIM 0.9955 0.9950
1˚ Rotation 0.8817 46.4286
BER 0 0
NC 0.9724 0.9873 QF = 10 0.7924 0
QF = 50 JPEG Compression 0.9388 0
QF = 90 0.9796 0
0.001 0.9687 0
Speckle noise
0.005 0.7564 0
Table 7
[20 20 400 480] Cropping 0.3586 45.2381
Evaluation of imperceptibility of proposed technique through subjective measure.
Gain Factor Visual quality of watermarked image [1 1] Median filter 0.9860 0

Proposed technique with Proposed technique with Histogram equalization 0.7223 0

chaotic-LZW Hyper chaotic-LZW
0.001 Excellent Excellent
0.05 Excellent Excellent
0.1 Very good Very good Table 10
0.5 Good Good Performance of NC and BER value for different attacks using Hyperchaotic-LZW
0.7 Acceptable Acceptable technique.
1 Acceptable Poor Attack Noise density NC (image) BER (text)
5 Poor Very Poor
0.0001 0.9836 0
Salt and pepper noise 0.0005 0.9769 0
0.001 0.9687 0
0.0001 0.9819 0
Gaussian noise
0.0005 0.9545 0
attacks. NC is also acceptable for most of the considered attack except Rotation 1˚ 0.9221 46.4286
for cropping.
QF = 10 0.8952 0
Table 11 shows a comparative assessment of our technique along JPEG Compression QF = 50 0.9510 0
with the amount of increase in the robustness when compared with [4] QF = 90 0.9809 0
method with similar existing approach. Several attacks were taken 0.001 0.9800 0
Speckle noise
into considerations, concluding that our technique is superior for eight 0.005 0.9014 0
considered attacks. A significant improvement in robustness can be Cropping [20 20 400 480] 0.5082 45.2381
experienced. The fusion of dual watermarking with cryptography, Error Median filter [1 1] 0.9819 0
Correcting Codes, compression and implementation of ETC makes our Histogram equalization 0.8716 0
technique more robust and secure, making it more superior. Maximum
improvement of 48% could be seen in histogram equalization attack.
Significant amount of increase in robustness can also be observed
against salt and pepper noise and cropping attacks. benchmark attacks. Further, some advanced techniques such as block
chain, deep learning or machine learning, better error correction code
6. Conclusion can be included to implement blind and more secure watermarking
system.
In this paper, an improved non-blind, robust, imperceptible and
Declaration of competing interest
secure watermarking technique is developed for concealing multi-
watermarks using DWT-SVD with hamming code. To reduce the pos- The authors declare that they have no known competing finan-
sibility of channel noise, the text watermark is encoded with hamming cial interests or personal relationships that could have appeared to
code prior to concealing into the cover image. By applying the concepts influence the work reported in this paper.
of encryption-compression on watermarked image, the confidentiality
and compression performance of the proposed method is improved. Acknowledgments
The result demonstration has shown good performance in terms of dis-
tortion (imperceptibility), robustness against various attacks, security The authors thanks the potential reviewer’s for their constructive
and compression ratio. In our future work, DWT and SVD may replace comments and suggestions that greatly contributed to improving this
with some more appropriate transform domain techniques to withstand article.

Table 8
Performance analysis for various length of text watermark.
Text watermark ETC technique PSNR (in dB) NC BER SSIM CR
Chaotic-LZW 36.1058 0.9835 0 0.996157 0.9085
‘‘Ashima NIT01’’
Hyper chaotic-LZW 34.2674 0.9668 0 0.996074 0.9466
Chaotic-LZW 36.0640 0.9820 0 0.996119 0.9084
‘‘BRTPA0411023’’
Hyper chaotic-LZW 34.2674 0.9672 0 0.996052 0.9466
Chaotic-LZW 36.0082 0.9833 0 0.996069 0.9084
‘‘NIT Patna’’
Hyper chaotic-LZW 34.2674 0.9692 0 0.995996 0.9466

79
A. Anand and A.K. Singh Computer Communications 152 (2020) 72–80

Table 11
Performance of proposed scheme against technique [4] method.
Attacks NC [4] NC (Proposed technique with NC (Proposed technique with
chaotic-LZW technique) Hyper chaotic-LZW technique)
Value Percentage Value Percentage
improvement improvement
JPEG compression (QF-60) 0.9325 0.9904 6.20 0.9839 5.51
Gaussian noise with mean = 0, Var = 0.05 0.3150 0.3226 2.41 0.3215 2.06
Salt and pepper noise (Density = 0.001) 0.7553 0.7751 2.62 0.9687 28.25
Scaling factor 2 0.7075 0.7075 0 0.7157 1.16
Histogram-Equalization 0.5880 0.7223 22.82 0.8716 48.23
Median filtering [2,2] 0.9116 0.9457 3.74 0.9438 3.53
Gaussian LPF with standard deviation = 0.6 0.8780 0.7265 0.8045
Sharpening mask with threshold = 0.1 0.6073 0.6763 11.36 0.6506 7.12
Cropping attack 0.7451 0.9337 25.31 0.9279 24.53

References [23] D.S. Chauhan, A.K. Singh, B. Kumar, J.P. Saini, Quantization based multiple
medical information watermarking for secure e-health, Multimedia Tools Appl.
[1] B. Hassan, R. Ahmed, B. Li, O. Hassan, An imperceptible medical image 78 (4) (2019) 3911–3923.
[24] A. Soualmi, A. Alti, L. Laouamer, A blind image watermarking method for
watermarking framework for automated diagnosis of retinal pathologies in an
personal medical data security, in: International Conference on Networking and
eHealth arrangement, IEEE Access 7 (2019) 69758–69775.
Advanced Systems (ICNAS), 2019.
[2] A.B. Houshyari, M. Bahadorani, M. Tootoonchi, J.J. Gardiner, R.A. Pena, P.
[25] E. Mansoori, S. Soltani, A new semi-blind watermarking algorithm using ordered
Adibi, Information and communication technology in medical education: an
hadamard transform, J. Imaging Sci. 64 (4) (2016) 204–214.
experience from a developing country, J. Pak. Med. Assoc. 62 (3) (2012).
[26] S.M. Mousavi, A. Naghsh, A.A. Manaf, S.A.R. Abu-Bakar, A robust medical image
[3] A.K. Singh, B. Kumar, G. Singh, A. Mohan, Medical Image Watermarking: Tech- watermarking against salt and pepper noise for brain MRI images, Multimedia
niques and Applications, in: Book Series on Multimedia Systems and Applications, Tools Appl. 76 (2017) 10313–10342.
Springer, USA, 2017. [27] A. Ustubioglu, G. Ulutas, A new medical image watermarking technique with
[4] A.K. Singh, B. Kumar, M. Dave, A. Mohan, Robust and imperceptible dual finer tamper localization, J. Digit Imaging 30 (2017) 665–680.
watermarking for telemedicine applications, Wirel. Pers. Commun. 80 (4) (2014) [28] N. Falgun Thakkar, V. Kumar Srivastava, A blind medical image watermarking:
1415–1433. DWT-SVD based robust and secure approach for telemedicine applications,
[5] https://www.experian.com/blogs/ask-experian/identity-theft-statistics/. Multimedia Tools Appl. 76 (2017) 3669–3697.
[6] C. Kumar, A.K. Singh, P. Kumar, A recent survey on image watermarking [29] A. Singh, M. Dave, A. Mohan, Hybrid technique for robust and imperceptible
techniques and its application in e-governance, Multimedia Tools Appl. 77 (2017) multiple watermarking using medical images, Multimedia Tools Appl. 75 (14)
3597–3622. (2015) 8381–8401.
[7] M.T. Ahvanooey, Q. Li, H.J. Shim, Y. Huang, A comparative analysis of infor- [30] H.V. Singh, A. Rai, Medical image watermarking in transform domain, Adv.
mation hiding techniques for copyright protection of text documents, Hindawi Intell. Syst. Comput. 48 (2018) 5–493.
Secur. Commun. Netw. 2018 (2018) 1–22. [31] Ali Al-Haj, Combined DWT-DCT digital image watermarking, J. Comput. Sci. 3
[8] S. Thakur, A.K. Singh, S.P. Ghrera, M. Dave, Watermarking techniques and its (9) (2007) 740–746.
applications in tele-health: A technical survey, in: Cryptographic and Information [32] D. Ravichandran, R. Nimmatoori, M.G. Ahamad, Mathematical representations
Security Approaches for Images and Videos by S. Ramakrishnan Chapter -17, of 1D, 2D and 3D wavelet transform, Image Coding 5 (3) (2016) 2319–2526.
[33] A.K. Singh, M. Dave, A. Mohan, Hybrid technique for robust and imperceptible
2018, pp. 467–511.
image watermarking in DWT- DCT-SVD domain, Nat. Acad. Sci. Lett. 37 (4)
[9] K. Sahu, P. Malviya, A survey on robust image watermarking techniques and
(2014) 351–358.
different applications, Int. J. Sci. Res. Eng. Trends 4 (4) (2018) 656–659.
[34] N. Singh, S. Joshi, S. Birla, Color image watermarking with watermark authen-
[10] J. Liu, J. Li, J. Ma, N. Sadiq, Y. Ai, FDCT and perceptual hash-based wa-
tication against false positive detection using SVD, in: International Conference
termarking algorithm for medical images, Smart Innov. Syst. Technol. (2019)
on Sustainable Computing in Science, Technology and Management (SUSCOM),
157–168.
2019.
[11] T.K. Araghi, A.A. Manaf, An enhanced hybrid image watermarking scheme for [35] K.A. Navas, M.C. Ajay, M. Lekshmi, Tampy S. Archana, M. Sasikumar, DWT-DCT-
security of medical and non-medical images based on DWT and 2-D SVD, Future SVD based watermarking, in: 3rd International Conference on Communication
Gener. Comput. Syst. 101 (2019) 1223–1246. Systems Software and Middleware and Workshops (COMSWARE ’08), 2008.
[12] Q. Dai, J. Li, U.A. Bhatti, Y.-W. Chen, J. Liu, SWT-DCT-based robust [36] C.-C. Chang, K.-N. Chen, C.-F. Lee, L.-J. Liu, A secure fragile watermarking
watermarking for medical image, Smart Innov. Syst. Technol. (2019) 93–103. scheme based on chaos-and-hamming code, J. Syst. Softw. 84 (9) (2011)
[13] P. Aparna, P.V.V. Kishore, Bio -metric based efficient medical image wa- 1462–1470.
termarking in ehealthcare application, IET Image Process. 13 (3) (2018) [37] I. Lee, M. Sullivan, E. Krimer, D.W. Kim, M. Basoglu, D.H. Yoon, L. Kaplan, M.
421–428. Erez, Survey of error and fault detection mechanisms, Technical Report, LPH
[14] M. Kumar, A. Vaish, An efficient encryption-then-compression technique for Group, Department of Electrical and Computer Engineering, The University of
encrypted images using SVD, Digit. Signal Process. 60 (2017) 81–89. Texas at Austin, 2012.
[15] T. Chuman, W. Sirichotedumrong, H. Kiya, Encryption-then-compression systems [38] P. Aparna, P.V.V. Kishore, Biometric-based efficient medical image watermarking
using grayscale-based image encryption for JPEG images, IEEE Trans. Inf. in E-healthcare application, IET Image Process. 13 (3) (2019) 421–428.
Forensics Secur. 14 (8) (2018) 1–11. [39] J. Sun, J. Lang, C. Miao, N. Yang, S.S. Wang, A digital watermarking algorithm
[16] I.J. Cox, G. Doërr, T. Furon, Watermarking is not cryptography, in: Y.Q. Shi, B. based on hyperchaos and discrete fractional fourier transform, in: 2012 5th
Jeon (Eds.), Digital Watermarking. IWDW 2006, in: Lecture Notes in Computer International Congress on Image and Signal Processing, 2012.
Science, vol. 4283, 2006, pp. 1–15. [40] S. Zhang, T. Gao, L.L. Gao, A novel encryption frame for medical image with
[17] P. Khare, A.K. Verma, V.K. Srivastava, Digital image watermarking scheme in watermark based on hyperchaotic system, Math. Probl. Eng. 2014 (2014) 1–11.
[41] C.-C. Chen, C.-C. Chang, High-capacity reversible data-hiding for LZW codes, in:
wavelet domain using chaotic encryption, in: Students Conference on Engineering
Second International Conference on Computer Modeling and Simulation, 2010,
and Systems, 2014.
pp. 3–8.
[18] A. Boonyapalanant, M. Ketcham, M. Piyaneeranart, Hiding patient injury in-
[42] G. Badshah, S.-C. Liew, J.M. Zain, M. Ali, Watermark compression in med-
formation in medical images with QR code, Adv. Intell. Syst. Comput. (2019)
ical image watermarking using Lempel–Ziv-Welch (LZW) lossless compression
258–267.
technique, J. Digit. Imaging 29 (2) (2015) 216–225.
[19] R. Mothi, M. Karthikeyan, Protection of bio medical iris image using [43] M. Stabno, R. Wrembel, RLH: Bitmap compression technique based on run-length
watermarking and cryptography with WPT, Measurement 136 (3) (2019) 67–73. and Huffman encoding, Inf. Syst. 34 (4–5) (2009) 400–414.
[20] P. Murali, V. Sankaradass, An efficient ROI based copyright protection scheme [44] Y. Wu, J.P. Noonan, S. Agaian, NPCR and UACI randomness tests for image
for digital images with SVD and orthogonal polynomials transformation, Optik encryption, Cyber J.: Multidiscip. J. Sci. Technol. J. Sel. Areas Telecommun.
170 (10) (2018) 242–264. (JSAT) (2011) 31–38.
[21] L. Bao, Y. Zhou, Image encryption: generating visually meaningful encrypted [45] S.R. Kodituwakku, U.S. Amarasinghe, Comparison of lossless data compression
images, Inform. Sci. 324 (2015) 197–207. algorithms for text data, Indian J. Comput. Sci. Eng. 1 (4) (2010) 416–426.
[22] Sonika C. Rathi, Vandana S. Inamdar, Medical images authentication through [46] https://medpix.nlm.nih.gov/home.
watermarking preserving ROI, Health Inform. - Int. J. (HIIJ) 1 (1) (2012) 27–42.

80