5388 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 8, NO.

OTE SENSING, VOL. 8, NO. 11, NOVEMBER 2015

Secure Dissemination and Protection of Multispectral

Images Using Crypto-Watermarking
Sangita Zope-Chaudhari, Parvatham Venkatachalam, and Krishna Mohan Buddhiraju

Abstract—Copyright protection of multispectral images and its watermarked multispectral images at dissemination. Although
integrity become one of the key problems in spatial informa- standard encryption algorithms like AES, DES, and RSA [7]
tion service due to rapid development of communication network. are available, they cannot be utilized for multispectral images
This paper depicts a crypto-watermarking scheme by combin-
ing watermarking and encryption to protect copyright of mul- due to high-volume data and nature of multispectral images. A
tispectral images and to provide security to the watermarked single key is not feasible for encryption of multispectral images.
image at dissemination level. We have proposed wavelet-based For multiple keys, key management for such a high-volume
watermarking and multiplicative-transposition-based cipher for data is critical. Moreover, these algorithms take more time
encryption. Experimental results show that the proposed algo- for encryption as well as decryption. Crypto-watermarking, a
rithm has not only good robustness against encryption attacks, but
also has transparency, strongness, large data hiding capacity, cor- combination of encryption and watermarking can be the best
rect extraction of watermark, and strong robustness against JPEG solution to provide total security for multispectral images.
lossy compression, filtering, and noise. Moreover, the classification Xu et al. [8] proposed a scheme by combining watermarking
of original and watermarked data is performed and it is observed and encryption. They tried to embed watermark at decryp-
that the proposed watermarking scheme leads to less misclassifica- tion of the image. This scheme does not provide total security
tion and it meets all the requirements for copyright protection as
well as the security of multispectral images in an effective manner. as the encrypted image does not carry watermark. Jiang and
Xu [9] depicted commutative encryption and watermarking
Index Terms—Classification, crypto-watermarking, discrete solution for remote sensing images. They utilized spatial scram-
wavelet transform, encryption, multispectral image.
bling based on Arnold scrambling for encryption using a single
key for whole remote sensing image. If a key is compro-
I. I NTRODUCTION
mised, the whole encryption system fails. Jiang et al. [10]

D UE TO advancement in technology, it becomes easy to

digitize text, images, videos, etc. Digital data can be
accessed and shared easily with the help of Internet. But this
tried to integrate encryption and watermarking using orthog-
onal decomposition of remote sensing images. This scheme has
severe effects on edges of remote sensing images and needs
leads to rampant misuse of digital data. The acquisition of some preprocessing at watermarking to retain edge informa-
multispectral images as well as preprocessing of multispectral tion. In all existing systems, complete security analysis as well
images are cost- and manpower-intensive tasks. Therefore, it as watermark robustness analysis are not done.
is important to protect the ownership rights of the data owner. This paper focuses on complete security protection for sensi-
Digital watermarking serves as a solution over the above said tive multispectral images by combining robust wavelet-based
problem. Multispectral images are used in various applications watermarking and encryption based on simple and efficient
including defense, which are related to national security. This cipher. The proposed watermarking system satisfies all mul-
makes multispectral images highly sensitive and its security tispectral image watermarking requirements. This approach
needs urgent attention. Moreover, the dissemination of such is suitable for secure dissemination and protection of large
sensitive images over the publicly accessed Internet is prone size multispectral images by ensuring security as well as the
to information leakage. Therefore, certain precautions have robustness of the whole crypto-watermarking technique.
to be adopted in handling the transfer of such sensitive and This paper is structured as follows. Section II gives brief
confidential images over the Internet. overview of discrete wavelet transform. Proposed crypto-
Security of multispectral images contains two aspects: secu- watermarking scheme is presented in Section III. Section IV
rity at storage and security at usage. These aspects can be very describes error analysis measures. Experimental results are
well handled by technologies like watermarking and encryp- discussed in Section V. Conclusion is drawn in Section VI.
tion. There is plenty of literature available for watermarking of
multispectral images both in spatial and frequency domain [1]–
II. D ISCRETE WAVELET T RANSFORM
[6]. However, none of them have considered security aspect of
The wavelet transform is based on wavelets. Wavelets are
Manuscript received September 30, 2014; accepted August 13, 2015. Date of
publication September 22, 2015; date of current version January 18, 2016. mathematical function representing scaled and translated copies
The authors are with the Indian Institute of Technology (IIT) Bombay, of a finite-length waveform called mother wavelet. It helps
CSRE, Mumbai 400076, India (e-mail: [email protected]; [email protected]. to analyze the given image in different frequency compo-
ac.in; [email protected]).
nents at different resolution levels. Discrete wavelet transform
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. (DWT) is a multiresolution description of an image. DWT
Digital Object Identifier 10.1109/JSTARS.2015.2475169 splits the signal into high- and low-frequency coefficients.
1939-1404 © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BHILAI. Downloaded on August 31,2021 at 05:57:43 UTC from IEEE Xplore. Restrictions apply.
 ZOPE-CHAUDHARI et al.: SECURE DISSEMINATION AND PROTECTION OF MULTISPECTRAL IMAGES 5389

an attack on image, Arnold transform maximally disperses the

damaged bits into different parts of the watermarks resulting
into good robustness. Arnold transformation and its inverse for
2-D image are given by
    
x 11 x
= mod N (1)
y 12 y
     
x 2 −1 x
= mod N (2)
y −1 1 y

where N is the size of the image. When Arnold transform is

applied on image, it randomizes original organization of pix-
els. On enough iterations (Arnold key), it generates the original
image.
2) Watermark Embedding Process: The embedding algo-
Fig. 1. Proposed framework for secure dissemination and protection of multi- rithm uses a binary image as watermark and color multispectral
spectral images. image as a host image. Host multispectral image is decomposed
up to third level for watermark embedding. Low-frequency and
The high-frequency coefficients contain information about the high-frequency sub-bands are selected for watermark embed-
edge components, while the low-frequency coefficient is split ding to achieve acceptable performance of imperceptibility and
again into high- and low-frequency coefficients [11]. The 2-D robustness (Algorithm 1).
wavelet transform decomposes an image into lower resolution Algorithm 1. Watermark Embedding
approximation coefficients (LL) and detail coefficients such
as horizontal (HL), vertical (LH), and diagonal (HH) coeffi- Input: Host image (R), Binary Watermark (W ), Arnold Key
cients. Watermark embedding in low frequency (LL) increases (Ak )
robustness against compression, Gaussian noise, scaling, and Process:
cropping while watermarking in high frequency (HH) is robust (1) Scramble the watermark (length = N ) using Arnold’s
to histogram equalization and intensity adjustments. transformation using Arnolds key as given in
equation 1.
(2) Apply 2-D DWT on each channel of host multispec-
III. P ROPOSED C RYPTO -WATERMARKING S CHEME tral image up to 3 levels (LL3 , HL3 , LH3 , and HH3 ).
Multispectral image mostly includes satellite images and Select LL3 and HH3 coefficients for watermark
aerial photographs. The demand for these data has increased embedding.
dramatically due to the large number of applications capable (3) At third level, calculate watermark strength (alpha) as a
to use them. Digital watermarking and encryption are used function of wavelet coefficients.
to achieve copyright protection and security of multispectral   
images at dissemination level. A wavelet-based algorithm is alpha = mean mean Ckch (i, j)
developed for copyright protection. Simple and strongly secure where i, j = 1, .., n ch = R, G, B k = LL3 , HH3
encryption based on multiplicative and two-stage transposition
cipher is used to provide security at the transmission level. (4) Watermark is embedded in selected wavelet coefficients
Fig. 1 depicts the flowchart of the proposed scheme. as:

C̃kch (i, j) = Ckch (i, j) + alpha × W (l, m)

A. Wavelet-Based Watermarking
where i, j = 1..n l, m = 1 . . . N ch = R, G, B
Although, general purpose image watermarking schemes k = LL3 , HH3
can be applied to multispectral image, these schemes have
to fulfill all the specific requirements for multispectral image (5) Apply inverse DWT to obtain watermarked multispectral
watermarking. The requirements are: 1) the watermarking tech- image.
nique should not distort certain specific areas in the image;
2) less distortion; 3) good robustness against attacks; 4) invis- Output: Watermarked image (R̃)
ible watermarking scheme; and 5) retention of classification
accuracy [12].
1) Watermark Preprocessing: To enhance the robustness 3) Watermark Extraction Process: Watermark extraction is
and security of the watermarking scheme, it is always desirable the inverse procedure of embedding. A watermark is extracted
to apply some transformation on watermark before embed- by applying DWT on both original data and watermarked data.
ding it into host data. Arnold transform is a simple, easy to Low-frequency coefficients of both are compared to detect
implement, and iterative permutation method [13]. In case of watermark W (Algorithm 2).

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BHILAI. Downloaded on August 31,2021 at 05:57:43 UTC from IEEE Xplore. Restrictions apply.
 5390 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 8, NO. 11, NOVEMBER 2015

Algorithm 2. Watermark Extraction

Input: Watermarked image (R̃), Host image (R), Arnold Key
(Ak )
Process:
(1) Using 2-D DWT, perform third level decomposition of the
watermarked multispectral image.
(2) Calculate alpha from selected coefficients (LL3 , HH3 ) of
each band of host multispectral image (R).
(3) Extract the watermarks from LL3 and HH3 coefficients
and perform averaging to get scrambled watermark.
(4) Obtain the binary watermark (W ) by applying Arnold’s
inverse transformation using equation 2 and Arnold key
(AK ).

Output: Binary watermark (W )

B. Multiplicative and Transposition Cipher-Based Encryption

For encryption and decryption of multispectral images, we
are proposing multiplicative and transposition cipher (MTC)
based on symmetric key multiplicative affine cipher and two-
stage transposition cipher. Individually, they are vulnerable to
brute force, statistical, and cipher text-only attacks. However,
combination and proper cascading of these ciphers provide
more secure and strong cipher than the individual. The final
cipher generated using such cascading approach is so strong
that it is very difficult to break it. The detail flowchart depict-
ing sequence of operations and correct arrangement of keys
at each operation is shown in Fig. 2. We are applying MTC Fig. 2. MTC-based encryption scheme.
on multispectral image (red, green, and blue channel) of size
M × N and gray levels G. Even and odd row elements are expressed in terms of the logarithmic decibel scale. It is defined
multiplied by EKER and EKOR . Each row and column are using mean square error (MSE). For two M × N images
shifted by EKSR and EKSC . Even and odd column elements Io (x, y) and Iw (x, y), where Io is the original image and Iw
are multiplied by EKEC and EKOC . Multiplier parameters is watermarked image, MSE and PSNR are given as
EKER , EKOR , EKEC , and EKOC are relatively prime to
M
 −1 N
 −1
G, whereas 0 < EKSR < M and 0 < EKSC < N . 2
M SE = Io (i, j)− Iw (i, j) (3)
Inverse MTC has keys DKEC , DKOC , DKSR , DKSC ,
i=0 j=0
DKER , and SKOR for even columns, odd columns,  
shifting rows, shifting column, even rows, and odd M AXI2w
P SN R = 10.log10 . (4)
rows, respectively, satisfying the conditions: EKER ∗ M SE
DKER = 1(mod G); EKOR ∗ DKOR = 1(mod G);
EKEC ∗ DKEC = 1(mod G); and EKOC ∗ DKOC = To evaluate the similarity between original watermark (wk)
1(mod G). For RGB multispectral image, we have (3 × 6)! and extracted watermark (w k), normalized correlation (NC) is
options for key arrangement. At decoder side, both the correct calculated. A perfect similarity gives a coefficient of one. Thus,
sequence of operations and correct keys should be available to the higher value of correlation coefficient indicates maximum
get the correct decryption; otherwise decoder cannot recover similarity. NC is calculated using
the original image.
(wk · wk )
k
NC =
 2 . (5)
IV. E RROR A NALYSIS M EASURES
(wk )
The watermarking process always introduces small degra- k
dation in watermarked data. There are many error measures
available in literature for the evaluation of watermarked data Entropy E for an image I is given by (6). Here, Pr (αi ) is the
errors [14]. One of the widely used error measures is peak occurrence probability of the gray level αi and G is the total
signal-to-noise ratio (PSNR). It is the ratio between the max- number of gray levels. Entropy indicates randomness of the
imum possible power of a signal and the power of corrupting pixels in encrypted image. If each pixel has equal probability in
noise that affects the quality of its representation. Because encrypted image then entropy of 8 would corresponds complete
many signals have a very wide dynamic range, PSNR is usually randomness. Low value of entropy indicates less randomness
Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BHILAI. Downloaded on August 31,2021 at 05:57:43 UTC from IEEE Xplore. Restrictions apply.
 ZOPE-CHAUDHARI et al.: SECURE DISSEMINATION AND PROTECTION OF MULTISPECTRAL IMAGES 5391

Fig. 3. Original multispectral images. (a) Sat_Image1 (1280 × 1280). Fig. 5. Watermarking and encryption/decryption results. (a) Original multi-
(b) Sat_Image2 (1160 × 1080). (c) Sat_Image3 (800 × 800). (d) Sat_Image4 spectral image (Sat_Image1). (b) Watermarked Sat_Image1. (c) Encrypted
(960 × 840). (e) Sat_Image5 (760 × 640). (f) Sat_Image6 (640 × 640). watermarked Sat_Image1. (d) Decrypted watermarked Sat_Image1.
(e) Decrypted watermarked Sat_Image1 with correct sequence of operations
and incorrect keys. (f) Decrypted watermarked Sat_Image1 with correct keys
and incorrect sequence of operations.

Secure image cryptosystems need high key sensitivity so that

Fig. 4. Original watermark, scrambled watermark by Arnold scrambling, and the image cannot be decrypted correctly even if there is a very
extracted watermark with NC = 1.
small change in correct keys. Keys used in multiplicative cipher
and can be used by attackers to launch an attack on encrypted are more sensitive as inverse of these keys only exists if it is
image [15] relatively prime to gray levels of an image. Original image can-
not be obtained even if a very small change occurs in these
G
 keys. The sequence of operations in MTC is also sensitive, as
E(I) = pr (αi )log2 (pr (αi )) . (6) a small sequence change can result into an incorrect decrypted
i=1 multispectral image.
Fig. 5(a)–(c) shows original, watermarked, and encrypted
V. E XPERIMENTAL A NALYSIS multispectral image (RGB) of size 1280 × 1280 × 3
We have tested the performance of the proposed system on with following keys. MTC encryption keys are: for
different multispectral images of varying sizes (Fig. 3). All red, EKER_R = 26, EKOR_R = 67, EKEC_R = 133,
the experiments are conducted on MATLAB on a machine EKOC_R = 91, EKSR_R = 797, and EKSC_R = 19; for
with 2.27 GHz Core 2 Duo processor and 4 GB RAM. The green, EKER_G = 41, EKOR_G = 109, EKEC_G = 118,
watermark of size 25 × 50 pixels is used for watermarking. To EKOC_G = 14, EKSR_G = 109, and EKSC_G = 71; and
improve the robustness of the watermarking scheme, original for blue, EKER_B = 59, EKOR_B = 98, EKEC_B = 73,
watermark is scrambled with Arnold’s transformation. In the EKOC_B = 137, EKSR_B = 567, and EKSC_B = 911.
experiment, we have taken 27 as an Arnold’s key for the water- Fig. 5(d) shows correctly decrypted multispectral image with
mark of size 25 × 50 as shown in Fig. 4(a), which produces exact keys and correct sequence of operations in MTC. MTC
the scrambled watermark as shown in Fig. 4(b). The extracted parameters are: for red, DKER_R = 206, DKOR_R = 118,
watermark with NC = 1 is shown in Fig. 4(c). We have consid- DKEC_R = 232, DKOC_R = 241, DKSR_R = 797, and
ered “Haar” wavelet in our experiment. For encryption, MTC is DKSC_R = 19; for green, DKER_G = 56, DKOR_G = 124,
proposed which combines multiplicative cipher with two-stage DKEC_G = 67, DKOC_G = 164, DKSR_G = 109,
transposition cipher. Total 36 keys are used in encryption and and DKSC_G = 71; and for blue, DKER_B = 134,
decryption procedure. DKOR_B = 242, DKEC_B = 7, DKOC_B = 188,
DKSR_B = 567, and DKSC_B = 911. Fig. 5(e) shows
incorrect decrypted multispectral image with incorrect keys
A. Security Analysis and correct sequence of operations in MTC. Here, only
A good encryption algorithm should robust against all kinds three keys changed (DKOC_R = 240, DKOC_G = 163,
of brute force, cryptanalytic, and statistical attacks. The his- and DKOC_B = 187). Fig. 5(f) shows incorrect decrypted
togram of encrypted image should be uniform to avoid statisti- multispectral image with exact keys and incorrect sequence of
cal attacks. Similarly, key space must be large enough to avoid operations in MTC.
brute force attacks. In this section, the robustness of the pro- A small change in encryption or decryption keys results into
posed encryption algorithm is presented. It has been observed incorrect decryption. Also, sequence of operations should be
that the proposed algorithm is robust against all the mentioned correct to get the correct decrypted image.
attacks. 2) Robustness Against Known-Plaintext Attack and Chosen
1) Key Sensitivity Analysis: In MTC, correct encryption Cipher Text Attack: These are basic attacks applied to the
and decryption are highly sensitive to the utilized key space. encryption algorithm. In known plaintext attack, the attacker

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BHILAI. Downloaded on August 31,2021 at 05:57:43 UTC from IEEE Xplore. Restrictions apply.
 5392 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 8, NO. 11, NOVEMBER 2015

TABLE II
PSNR R ESULTS

Fig. 6. RGB channel histogram of original and encrypted multispectral image

(Sat_Image1). (a)–(c) RGB channel histogram of original multispectral image.
(d)–(f) RGB channel histogram of encrypted multispectral image. TABLE III
T IME A NALYSIS OF WATERMARKING AND E NCRYPTION /D ECRYPTION
TABLE I
E NTROPY OF O RIGINAL , WATERMARKED , AND E NCRYPTED /D ECRYPTED
M ULTISPECTRAL I MAGES

has knowledge of some part of plaintext and corresponding

cipher text. The attacker tries to find the secret keys by cor-
relating plaintext and cipher text. In chosen cipher text attack,
the attacker tries to find plaintext by applying unknown keys
on randomly selected cipher text. MTC is robust against these
attacks as it requires correct keys as well as the correct sequence
of operations. Even if an attacker knows correct keys, but does
not know the exact sequence of operations, image cannot be
decrypted correctly.
3) Histogram Analysis: The encryption algorithm is strong
if it possesses good confusion and diffusion properties.
Histogram analysis is used to demonstrate confusion and dif-
fusion of the proposed scheme. The color variations in RGB
channels of the original and encrypted image are represented
in terms of histogram in Fig. 6. It is observed that the his-
togram of the encrypted image is nearly uniformly distributed,
and significantly different from the respective histograms of the Fig. 7. Attacked watermarked images. (a) JPEG compression (50%).
(b) Median filter (5 × 5). (c) Average filter (5 × 5). (d) Histogram equalization.
original image. So the encrypted image does not provide clues
(e) Salt and pepper noise. (f) Gaussian noise. (g) Cropping (10%). (h) Scaling
to employ any statistical attack on the proposed scheme. (0.8). (i) Rotation (5◦ ).
As shown in Fig. 6, the occurrence probability of each of the
gray level is fairly equitably distributed leading to very high
entropy. Table I shows the entropy of different encrypted multi- of image decreases, the time required for both watermarking as
spectral images. Resultant entropy is nearly equal to 8 indicates well as encryption gets decreased.
that encrypted image is robust against statistical attacks.
4) Perceptual Security: PSNR is one of the most impor-
B. Watermarking Analysis
tant objective evaluation indexes for image quality. The smaller
value of PSNR means worse image quality and better per- Decrypted watermarked images are evaluated against imper-
ceptual security. The proposed scheme gives good encryption ceptibility, robustness to common image processing operations
perceptual security as it has very less value of PSNR between and classification accuracy. The resultant watermarked raster
original and encrypted as well as watermarked and encrypted data have high values of PSNR indicating less error and good
multispectral image (columns 2 and 3 in Table II). imperceptibility (Table II). Various geometric and nongeomet-
5) Time Analysis of Proposed System: For each of the mul- ric operations are performed on decrypted watermarked image
tispectral image of different size, time taken for watermarking (Sat_Image1) as shown in Fig. 7. Nongeometric operations
and encryption is recorded as shown in Table III. As the size include JPEG compression, median filtering, average filtering,

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BHILAI. Downloaded on August 31,2021 at 05:57:43 UTC from IEEE Xplore. Restrictions apply.
 ZOPE-CHAUDHARI et al.: SECURE DISSEMINATION AND PROTECTION OF MULTISPECTRAL IMAGES 5393

TABLE IV
NC C OEFFICIENTS FOR N ONGEOMETRIC ATTACKS

TABLE V
NC FOR G EOMETRIC ATTACKS

TABLE VI
C OMPARISON OF ROBUSTNESS

Fig. 8. Unsupervised classifications. (a) Original image. (b) Watermarked

image.

TABLE VII
C OMPARISON OF C LASSIFICATION ACCURACY

noise, and histogram equalization. Geometric operations are

cropping, rotation, and scaling.
Table IV shows the resistance of the proposed algorithm
against the listed nongeometric operations. It is noted from
the table that the proposed method resists operations such
as median filtering and average filtering with a mask up to
5 × 5. The extracted watermark from these operations can Robustness of proposed algorithm is better compared to exist-
be recognized clearly. It is robust against Gaussian, salt and ing algorithm for all attacks except rotation.
pepper as well as Poisson noise. We have also checked robust- Watermarking process modifies an image pixel brightness
ness against geometric operations like scaling, cropping, and value which has a straight impact on accuracy of image clas-
rotation. Table V shows NC between original and extracted sification algorithms. Unsupervised classification (ISODATA)
watermarks. The proposed algorithm is robust against scaling, is performed on original and watermarked multispectral image.
cropping, and rotation attack. Fig. 8(a) and (b) shows the classification of original and
The performance of the proposed algorithm is compared watermarked multispectral image, respectively.
with Fourier-based watermarking proposed by Barni et al. Table VII shows the result of classification with six clusters.
[16]. Table VI shows the comparison of these two algorithms. It shows the total number of pixels (%) that were assigned to the

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BHILAI. Downloaded on August 31,2021 at 05:57:43 UTC from IEEE Xplore. Restrictions apply.
 5394 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 8, NO. 11, NOVEMBER 2015

same class in the watermarked multispectral image as they were [6] B. Ziegeler, H. Tamhankar, J. Fowler, and L. Bruce, “Wavelet-Based
in a classification of the original multispectral image. It also watermarking of remotely sensed imagery tailored to classification per-
formance,” in Proc. IEEE Workshop Adv. Techn. Anal. Remotely Sensed
shows the percentage of pixels that have changed the classes. Data, 2003, pp. 259–262.
[7] B. Schneier, Applied Cryptography. Hoboken, NJ, USA: Wiley, 1995.
[8] Y. Xu, Z. Xu, and Y. Zhang, “Content security protection for remote
VI. C ONCLUSION sensing images integrating selective content encryption and digital fin-
gerprint,” J. Appl. Remote Sens., vol. 6, no. 1, p. 063505, 2012.
In this paper, we have proposed crypto-watermarking, a com- [9] L. Jiang and Z. Xu, “Commutative encryption and watermarking for
remote sensing image,” Int. J. Digital Content Technol. Appl., vol. 6,
bination of watermarking and encryption to provide copyright no. 4, pp. 197–205, 2012.
protection for multispectral images and secure delivery of copy- [10] L. Jiang, Z. Xu, and Y. Xu, “A new comprehensive security protection
righted multispectral images. During transmission, encryption for remote sensing image based on the integration of encryption and
watermarking,” in Proc. IEEE Int. Geosci. Remote Sens. Symp., 2013,
can be used to prevent information leakage and protocol pp. 2577–2580.
attacks. Ownership can be proved later as invisible water- [11] S. Mallat, “The theory for multiresolution signal decomposition: The
mark is retained in the multispectral image. Ownership cannot wavelet representation,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 11,
no. 7, pp. 654–693, 1989.
be proved until and unless the original multispectral image [12] M. Barni, F. Bartolini, V. Cappellini, E. Magli, and G. Olmo,
is made available. It is observed that the proposed crypto- “Watermarking-based protection of remote sensing images:
watermarking approach satisfies the security of encryption, Requirements and possible solutions,” in Proc. SPIE Int. Symp.
Opt. Sci. Technol., vol. 4475, pp. 191–202, 2001.
the invisibility, robustness, and classification accuracy reten- [13] J. Zou and X. Tie, “Arnold transformation of digital image with two
tion of watermarking. Moreover, this algorithm survives all the dimensions and its periodicity,” J. North China Univ. Technol., vol. 12,
attacks having low-frequency as well as high-frequency char- pp. 10–14, 2000.
[14] M. Kutter and F. Petitcolas, “A fair benchmark for image watermark-
acteristics as we have utilized both low- and high-frequency ing systems,” in Proc. Int. Conf. Secur. Watermark. Multimedia Contents,
bands for watermarking. The same algorithm can further be 1999, pp. 226–239.
used for hyperspectral images security at storage as well as at [15] X. Wang, J. Zhao, and H. Liu, “A new image encryption algorithm based
on chaos,” Opt. Commun., vol. 285, no. 5, pp. 562–566, 2012.
dissemination. [16] M. Barni, F. Bartolini, E. Magli, O. Gabriello, and R. Zanini, “Copyright
protection of remote sensing imagery by means of digital watermarking,”
in Proc. SPIE Int. Symp. Remote Sens., 2001 vol. 4550, pp. 565–576.
R EFERENCES
[1] M. Barni, F. Bartolini, V. Cappellini, E. Magli, and G. Olmo, “Near- Sangita Zope-Chaudhari received the M.E. degree in computer engineering
lossless digital watermarking for copyright protection of remote sensing from Mumbai University, Mumbai, India, in 2008.
images,” in Proc. IEEE Int. Geosci. Remote Sens. Symp., 2002, pp. 1447– She is currently a research scholar at CSRE, IIT Bombay, Mumbai, India.
1449. Her research interests include digital image processing, advanced database and
[2] Y. Chauhan, P. Gupta, and K. Majumder, “Digital watermarking of information systems, and information security techniques.
satellite images,” in Proc. 3rd India Conf. Comput. Vis. Graph. Image
Process., 2002, pp. 235–240. Parvatham Venkatachalam received the M.Sc. and Ph.D. degrees in mathe-
[3] T. Hemalatha, V. Joevivek, K. Sukumar, and K. Soman, “Robust water- matics from IIT Bombay, Mumbai, India, in 1972 and 1978, respectively.
marking of remote sensing images without the loss of spatial infor- She is currently a Professor with CSRE, IIT Bombay, India. Her research
mation,” in Proc. 10th ESRI India User Conf., 2009, vol. 1, no. 2, interests include geographic information systems, spatial decision support
pp. 1–8. systems, spatial data structure, spatial data mining, and warehousing.
[4] B. Kumari and V. Rallabandi, “Modified patchwork-based watermarking
scheme for satellite imagery,” Signal Process., vol. 88, no. 4, pp. 891– Krishna Mohan Buddhiraju received the Ph.D. degree in electrical engineer-
904, 2008. ing from IIT Bombay, Mumbai, India, in 1991.
[5] P. Zhu and C. Chen, “A copyright protection watermarking algorithm for He is currently a Professor with CSRE, IIT Bombay, India. His research
remote sensing image based on binary image watermark,” Int. J. Light interests include high resolution image segmentation and classification, hyper-
Electron Opt., vol. 124, no. 20, pp. 4177–4181, 2013. spectral image analysis, and geographic information systems.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BHILAI. Downloaded on August 31,2021 at 05:57:43 UTC from IEEE Xplore. Restrictions apply.