International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 07 Issue: 04 | Apr 2020 www.irjet.net p-ISSN: 2395-0072

HIGH CAPACITY DUAL DATA HIDING IN ENCRYPTED IMAGES USING

MSB PREDICTION

Rugma M1, Umesha K,2

1M. Tech student, Electronics and Communication, Jawaharlal College of Engineering and
Technology, Kerala, India
2 Professor, Electronics and Communication, Jawaharlal College of Engineering and

Technology, Kerala, India

---------------------------------------------------------------------***----------------------------------------------------------------------
Abstract - In the past few years ,data privacy and protection is to ensure data privacy by fully or partially randomizing
of multimedia data are becoming more important. Because of the content of the original images. Reversible Data Hiding
(RDH) consist of embedding hidden data in a signal .It is
this, DHEI has become an investigative field. The multimedia
fundamental to reconstruct the original image with
data protection can be done with encryption or data hiding minimum number of errors or preferably none at all after its
algorithms. High capacity dual data hiding in encrypted extraction. The distortion of image may have a critical
images(DHEI) is an effective technique to embed data in impact in some strict areas, like in the medical world or
encrypted domain. An original image is encrypted using the military.
secret key, still possible to embed additional data without
Maetal [1] introduce a reversible data hiding methodology
knowing the original content of the image or the secret key.
for encrypted images to obtain an error free recovered
This secret message can be recovered and the initial image can images by reserving rooms before encryption with a
be extracted in the decoding phase. Recently, reversible data conventional data hiding algorithm where additional data is
hiding methods have been proposed with high capacity, but accommodated in the reserved room. In their method, image
these proposed methods do not allow a large amount of reconstruction and data extraction are free of errors.
embedding capacity. In this paper we propose a high capacity Zhangetal [2] proposed improved reversibility method
dual data hiding method based on MSB (most significant bit) before encrypting the image, room for data embedding is
vacated by shifting the histogram of estimating errors of
prediction. We suggest to hide two bit per pixel by pre- pixels .Image recovery and data retrieving are free of errors
processing the image to avoid prediction errors and, thereby, in their method. Although these two approaches greatly
improving the quality of the reconstructed image. We increase the embedding ability and reversibility, it may be
demonstrate two approaches: high capacity reversible data difficult to empty the content owner's space for data
hiding approach with correction of prediction errors(CPE- embedding, since reversible data hidden in encrypted image
HCRDH) and high capacity reversible data hiding with often allows the content owner to do nothing but image
encryption, and the service provider is expected to conduct
embeddedpredictionerrors(EPE-HCRDH).We have applied our
data embedding. Quianetal [3] proposed a separable
method on various images, every cases, our results are better reversible data hiding approach for encrypted image using a
than those results obtained from the current methods, both of N-nary data hiding method and histogram modification. At
embedding capacity and reconstructed image quality. the content owners side reserving room is no longer
required and a error free recovered image is obtained by
Keywords – CPE-HCRDH, EPE-HCRDH, High capacity dual
their method. These two reversible data hiding strategies for
data hiding, MSB prediction. encrypted images based on prediction error are introduced
to improve the problems. These methods can provide better
1. INTRODUCTION visual quality of loose recovered image and improved
reversibility. Further, the number of extracted incorrect bits
Digital image security plays an essential role in every field are significantly reduced by the join method.
particularly in computer network and has led to serious
security issues where authentication, confidentiality and In this paper we present a new high capacity reversible data
reputation are threaten by activities such as hacking, copying hiding scheme based on MSB prediction for encrypted
or malicious use of images. To ensure safe multimedia images. Two adjacent pixel values are very similar due to the
content transmission via the public communication channel local co-relation between a pixel and its neighbors in a clear
too major strategies have been developed: Encryption and image. Of this purpose it seems normal to estimate the pixel
Data hiding. During the encryption step without knowing the
value by using already decrypted previous one as in many
original content or the secret key used it is possible to image codding and compression schemes. For some
analyze or process the during the transmission or Archiving instances, however there are some errors. So the first step of
of the encrypted images. The aim of the encryption method
© 2020, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 3128
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 07 Issue: 04 | Apr 2020 www.irjet.net p-ISSN: 2395-0072

our method is to define all errors of prediction in the original  Calculate the value pred(i,j) from the previously
image and store this information in a error location binary scanned neighbors of p(i, j) which is considered as a
map. After that , we proposed two different approaches : The predictor during the decoding stage.
CPE-HCRDH (High capacity reversible data hiding with  Calculate the absolute difference between pred(i,j)
correction of prediction errors) and the EPEHCRDH (High and p(i, j) as well as pred(i,j) and inv(i,j).Record the
Capacity reversible data hiding with embedded prediction results as ∆ and ,so that
errors). The CPE-HCRDH approach consist of correcting the
prediction error prior to encryption. In order to remove all ∆=| pred(i,j)− p(i,j)|
the prediction errors the original image is pre-processed and (1) =| pred(i,j)− inv(i,j)|
then this image is encrypted based on the error location
map. EPEHCRDH approach, the original image is encrypted
 Compare the values of ∆ and .If ∆<
directly, after this encryption the location of prediction
errors is embedded (EPE).In both approaches during the there is no prediction error since the original value
data hiding Phase, the MSB value of each available pixel is of p(i, j) is equal to the predictor than the inverse
substituted by a bit of secret message in the encrypted value of the predictor itself. Otherwise an error
images. By the end of this process , the embedded data can occurs, and we store the information in a binary
be extracted without any errors and the original image can map of the error position.
be losslessly reconstructed by using MSB prediction.
2.2) Image Encryption
2. PROPOSED METHOD The original image is encrypted by using an encryption key
= (c, ), as shown in fig.2. The encryption is done by
This System uses a new reversible data hiding method for using a chaotic generator which generates a pseudo-random
encrypted images based on MSB prediction with a very high sequence. As a result a sequence of pseudo random bytes
capacity. We adapt the message to be inserted to highlight s(i,j) are obtained and the encrypted pixels p(i, j) can be
the problematic pixels without significantly reducing the calculated through exclusive-or (XOR) operation.
embedding capacity to avoid the prediction errors. The
encoding phase comprises of three steps: MSB prediction
(i,j)=s(i,j) p(i,j) (2)
error detection, joined MSB error consideration and
encryption, reversible data hiding by MSB substitution. This
Process is shown in Fig.1. There are three possible outcomes
in the decoding phase. The receiver has only the encryption
key they can only retrieve the original image, but not the
embedded message. On the contrary if they have only the
water marking key, they can just extract the hidden message.
Obviously when they are having both the encryption and Fig.2 Encryption step.
water marking keys, the recipient can extract the secret
message as well as reconstruct the original images.
2.3) Data Embedding
In this Phase its is possible to embed data in the encrypted
image without knowing either the encryption key used or
the original content of the image . The message to be
inserted is first encrypted by using the data hiding key ,in
order to prevent its detection after embedding in the marked
encrypted image. Next , the encrypted image pixels are
scanned from left to right , then from top to bottom (Scanned
line Order ) and the MSB of each available pixel substituted
by one bit , with 0 k < m n, of the secret message.

Fig. 1 Overview of general encoding method. (3)

2.1) Prediction error detection

2.4) Data Extraction and Image Recovery
We first analyze the original image content to detect all the
Since this method is separable it is possible to extract the
prediction errors in order to detect all the possible
prediction errors we first analyze the original image content. secret message and reconstruct the clear image separately
during the decoding phase. may be same as like the original
 Consider the current pixel p(i,j) where 0≤i<m and
image I itself or a processed image very close to the
0≤j<n and its inverse value which is inv(i,j)=(
p(i,j)+128)mod 256. • Note that the difference original image, based upon which approach is used. There
between those two values is equal to 128. are three possible outcomes:

© 2020, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 3129
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 07 Issue: 04 | Apr 2020 www.irjet.net p-ISSN: 2395-0072

1) the recipient has only the data hiding key ,

 (7)
2) the recipient has only the encryption key ,
3) the recipient has both keys.
An overview of the decoding method is presented in Fig. 3. A. CPE-HCRDH Approach

As shown in Fig.4, we first pre-process the original image in

the CPE-HCRDH approach (high-capacity reversible data
hiding approach with correction of prediction errors) to
avoid all the prediction errors so as to reconstruct the image
during the decoding phase. After this process, the pre-
processed image can be encrypted without any problems.
Each pixels of the encrypted image are marked with one bit
of the message during the embedding process.

1) Predictor Used
Fig.3 Overview of general decoding method we proposed to use the preceding pixels to estimate the
value of the current pixel. In this method(except for the first
If the recipient has only data hiding key ,the pixels of the row and the first column) we consider the average of the left
marked encrypted image are scanned in the scan line order and the top pixels as a predictor pred(i,j) for example:
and the MSB of each pixel are extracted in order to retrieve
the encrypted message. pred(i,j)=(p(i-1,j)+p(i,j-1))/2 (8)
= (i,j)/128 (4)
Where 0≤k<m×n refers to the index of the recovered bit in 2) Image Pre-Processing
the message. Then, corresponding plain text can be obtained We propose to pre-process the original image after
by using the data hiding key . prediction error detection phase in order to obtain an image
In the second scenario, if the recipient only without any prediction errors. For each problematic pixel,
has ,the image can be retrieved by we note the amplitude of the error and we determine the
value of the minimal pixel modification change required to
continuing as follows before the data hiding and encryption
prevent this error. Eq. 8 indicates the provision required to
steps:
prevent any errors in the prediction during the decoding
1) The encryption key is used to produce the s(i, j) phase:
sequence with pseudo-random bytes of m n.
2) The pixels of the marked encrypted image are scanned in | pred(i,j)-p(i,j)|<64 (9)
the order of scan line, and the seven LSB values are retrieved
by XORing the marked encrypted value (i,j) with the 3) Data Extraction and Image Recovery
associated binary sequence s(i,j) in the pseudo-random During the decoding process, the marked encrypted image
stream is scanned and the MSB of each pixel is simply extracted
j) = s(i,j) (i, j) (5) to extract the secret message. On the other hand, without
where represents the XOR operation. any modification the pre-processed image can be
3) The MSB value is predicted: reconstructed. To obtain the seven less significant bits we
 The value of the predictor pred(i,j) is computed by first decrypt the marked encrypted image and, then
using the values of the previously decrypted
predict the MSB value. The reconstructed image resembles
adjacent pixels.
very similar to the original one.
 The pixel value is determined with MSB = 0 and
with MSB = 1 and the variations between each of
these two values and pred(i,j) are computed. These
values are recorded as and :

| pred(i,j)- | (6) =| pred(i,j)-

| Fig.4 CPE-HCRDH approach encoding phase.

 The lowest value between and gives the pixel

value searched:

© 2020, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 3130
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 07 Issue: 04 | Apr 2020 www.irjet.net p-ISSN: 2395-0072

B. EPE-HCRDH Approach
The main objective of the EPE-HCRDH approach (high-
capacity, reversible data hiding method with embedded
prediction errors) is precisely to recreate the original image .
In this case, the payload may decrease a little because error
position information being stored. To highlight the
prediction errors, we adjust the information to be inserted
according to the error location binary map, created during
the of the prediction error detection process. The original
image is then encrypted and the error location information is Fig.5 prediction error highlighting
inserted in the encrypted image immediately afterwards. We
can hide only one bits of the secret message within the 3) Data Extraction and Image Recovery
available pixels during the data hiding step. Using this error The secret message can be extracted during the decoding
location information, the original image can be step.
reconstructed at the end of the decoding step, without any  The pixels of the marked-encrypted image are
apparent alteration, which is indicated by a PSNR which scanned in the order of the scan line and the MSB
tends to +∞. Fig.5 provides a regional scheme of this value is extracted and stored for each pixel. We
strategy. assume the extracted values are bits of the
embedded message before the first eight MSB series
1) Predictor Used equals 1.
We have two possible predictors in this scheme for each  When a sequence of this type is detected, it
pixel : the left pixel p(i; j - 1) and the top pixel indicates the start of the error sequence. Since the
p(i-1;j). The absolute difference with the current pixel p(i;j) next pixels are not marked during the data hiding
is calculated and the closest value is chosen to determine phase, pixels are scanned until the next sequence
which of those values is considered to be a predictor, where eight MSB are equal to 1, which indicates the
if |p(i-1,j)- p(i; j)|<| p(i,j-1)- p(i; j)|, end of the error series.
then, pred(i,j)=p(i-1,j), (10)  Repeat the process until the end of the image.
else, pred(i,j)= p(i,j-1),

2) Embedding of the Error Location Information

The location of the prediction errors is stored in the error
location binary map during prediction error detection. Then,
the original image I is encrypted. The encrypted image is
adapted before the embedding step to avoid prediction
errors. The encrypted image is then split into eight pixel
blocks and scanned by block in the scan line order. If,
according to the binary error location map, at least one
prediction error is found in a block the current block is Fig.6 EPE-HCRDH approach encoding phase.
surrounded by two flags by replacing the MSB of each pixel
in the previous and the subsequent blocks by one. In the 3. RESULT AND ANALYSIS
scurrent block,if there is a prediction error the MSB value of
a pixel is replaced by 1 and 0 if no error is detected as in Fig. Throughout this section we present the results obtained by
5.If there is no error in the current block, and it does not implementing our method using the CPE-HCRDH approach
serve as a flag, and then the eight pixels of this block are (high-capacity reversible data hiding approach with
used for data hiding. When error occurs in the two adjacent predictive error correction) and the EPE-HCRDH approach
blocks, the flag which indicates the end of the error sequence (high-capacity reversible data hiding approach with
and will be moved b without error until the next block. When embedded predictive error). We use two metrics with
the flags are used for more than one prediction error the loss complete reference which are peak-signal-to-noise ratio
of embedding capacity becomes less important. (PSNR) and structural similarity (SSIM) to evaluate the
reconstructed image quality in comparison to the original
one. We first applied our two approaches on different
original image also applied to files of different formats like
BMP, TIF. Fig. 5.1 shows the result obtained with CPE-
HCRDH and fig .5.2 with the EPE-HCRDH approach. In fig
5.1.a shows the original image and fig 5.2.a ,in white,
indicates the location of all the pixels with prediction
errors. We can observe that, we have neither the same

© 2020, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 3131
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 07 Issue: 04 | Apr 2020 www.irjet.net p-ISSN: 2395-0072

prediction errors, nor the same number of errors, since we image obtained during the final encoding step.(Fig .7.1.f)
indicates the reconstructed image after decoding process.
do not use the same predictor in these two approaches, but
they are in the same order of magnitude globally.
a) b)
a) b)

c) d)
c) d)

e) f)

e) f)

g)

Fig 7:Experiment using CPE-HCRDH approach a) original

image I b) error location, number of errors=117235 c)
error histogram d) encrypted image e)Marked
encrypted image f) reconstructed image

In the CPE-HCRDH approach (Fig. 7.1.a), indicates the

original image in which we are applying the process. Fig 8. Experiment using EPE-HCRDH approach a) original
(Fig. 7.1.b), indicates the original image pixels whose MSB image I b) error location, number of errors=117235 c)
would be badly predicted if we did not change their values error histogram (d) pre-processed image e) encrypted
during the pre-processing phase. The histogram in (Fig. image f)Marked encrypted image (g) reconstructed
7.1.c) demonstrates the prediction error distribution while image
using the CPE-HCRDH approach and shows the required
pixel modifications values to avoid all the prediction errors In the EPE-HCRDH approach (Fig. 8.2.a), indicates the
and (Fig. 7.1.d) represents the encrypted image using the original image in which we are applying the process.
encryption key. (Fig. 7.1.e) illustrates the marked encrypted (Fig. 8.2.b), these are original image pixels whose MSB would

© 2020, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 3132
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 07 Issue: 04 | Apr 2020 www.irjet.net p-ISSN: 2395-0072

be poorly predicted if we do not adapt their values during image content, while at the same time providing authenticity
or integrity check.
the pre-processing step. The histogram in (Fig. 8.2.c)
illustrates the distribution of the prediction errors while
using the EPE-HCRDH approach then shows the required ACKNOWLEDGEMENT
pixel value modifications to avoid all the prediction errors
and (Fig. 8.2.d) represents the pre-processed image. (Fig I am ineffably indebted to my guide Mrs. Vijitha G for her
8.2.e) represents the encrypted image using the encryption active guidance and encouragement to accomplish this
key.(Fig. 8.f) illustrates the marked encrypted images, assignment and Wikipedia for the correct source of
information.
obtained during the final encoding step.(Fig .8.2.g) indicates
the reconstructed image after decoding process. Note that in
both approaches the secret message is always extracted REFERENCES
without any errors.
[1] K.Ma,W.Zhang, X.Zhao, N.Yu,F.Li, Reversible data hiding
Table Ⅰillustrates the performance measurements obtained in encrypted images by reserving room before encryption,
for the test images. In most of the cases, when there is no IEEE Trans. Inf. ForensicsSecur.8(3)(2013)553–562.
prediction error the two approaches are fully reversible. In [2] W.Zhang, K.Ma, .Yu, Reversibility improved data hiding
this case, original images are recovered without any errors, in Encrypted images, SignalProcess.94(2014)118–127.
as a SSIM equal to 1. Furthermore, for low contrast images,
[3] Z.Qian, X.Han, X.Zhang, Separable reversible data hiding
the reconstructed image quality is high.
in encrypted images by n-nary histogram modification,
in:TheThird International Conference.
BEST WORST
CASE CASE [4] Pauline Puteaux, William Puech,An Efficient MSB
CPE- PSNR(Db) 44.5599 42.0128 Prediction- Based Method for High-Capacity Reversible Data
HCRDH SSIM 0.9997 0.9992 Hiding in Encrypted Images, Transactions on information
approach forensics and security.
EPE- PSNR(Db) + 32.7750 [5] W. Hong, T.-S. Chen, and H.-Y.Wu, “An improved
HCRDH SSIM 1 0.9952 reversible data hiding in encrypted images using side
approach match,” IEEE Signal Processing Letters, vol. 19, no. 4, pp.
Table I Performance measurements of the two 199–202, 2012.
approaches.
BIOGRAPHIES
4. CONCLUSIONS
Rugma M received the B.Tech
In this study, we proposed an efficient method of reversible degree from Calicut University,
data hiding in encrypted images with a very high embedding India, in 2016. She is currently a
capacity based on MSB prediction, which outperforms the final year M.Tech student of APJ
current state-of-the-art methods. From our knowledge this is Abdul Kalam Technological
one of the first approaches proposing to use MSB for an University. Her work has focused
RDHEI instead of LSB. Since MSB prediction in original on multimedia security in
domain is easier than LSB prediction and also degradation particular in image processing in
image quality is not a problem in the encrypted domain, we the encrypted domain.
are then able to have a very high capacity. By analyzing the
original image content, the prediction errors are identified Dr Umesha K a guide from the
and an error location binary map is created. The original department of ECE received his
image is slightly modified in the CPE-HCRDH approach in B.E from Karnataka University
order to remove all the prediction errors. After that, it is M.E from UVCE Bangalore and
possible to hide one bit per pixel by substituting all MSB in PhD from Magadh Unversity in
the image. Besides this maximal payload equal to 1 bpp, the the year 2010,his area of research
reconstructed image quality is high (SSIM close to 1).In the is biomedical Engineering and
EPE-HCRDH method, information about the location of the medical electronics has guided
prediction errors are stored in the encrypted image many projects in this area.he is
serving from 2012.
according to the error location binary map. By substituting
most of the MSB values in the encrypted image, a large
message can be embedded(close to payload 1 bpp) and the
original image can be losslessly recovered during the
decoding phase(PSNR =+∞ ).In addition, we have also seen
that the proposed scheme offers a good security level and
can be used to preserve the confidentiality of the original

© 2020, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 3133