SEMINAR REPORT ON Reversible Data Hiding in Encrypted

 

 

 

 

 

SEMINAR
REPORT ON

 

Reversible
Data Hiding in Encrypted Images

by
Reversible Image Transformation

 

SUBMITTED
TO

 Department of Computer Science &
Engineering

 

by

Chethan
R Bhat

 Registration number: 170948002

II
Sem M. Tech (CSIS)

 

 

Name
& Signature of guide

Mr.
Kishore B

Assistant Professor – Selection Grade

 

 

            

              Evaluator 1                                                                                        Evaluator
2

 

        Dr. Manjunath K N                                                                             Mr. Rohit Verma    

Assistant Professor-Senior
Scale                                                               Assistant Professor

 

 

CHAPTER 1

INTRODUCTION

 

Information
processing in the encrypted domain has attracted considerable research
interests. In numerous applications, for example, distributed computing and
assigned estimation, the content owner may need to transmit data to a remote
server for processing further if required. In few cases, the content proprietor
may not confide in the administration provider and requirements to encode the
information before transferring. Accordingly, the service supplier must have
the capacity to do the handling in the encrypted space. Scarcely any works have
been finished for information processing in an encryption space, for instance,
compressing of encoded pictures, including a watermark into the encrypted
picture, and reversibly hiding of information into the encrypted picture. Not
at all like powerful watermarking, reversible data hiding (RDH) is able to obtain
complete image reproduction and information extraction.

 

 

Reversible data
hiding (RDH) in pictures is the technique, in which the first cover picture may
be losslessly recuperated before which the hidden text has been extracted. Since
this imperative strategy is utilized as a part of medical domain, military area
and law crime scene investigation, where no loss of the cover image is
acknowledged. Since it keeps up the superb property that the cover image can be
recouped after inserted information is removed while securing the picture
substance’s privacy.

 

 

Reversible
Data Hiding (RDH) of encoded pictures has the great property that the first
cover picture can be losslessly reestablished after the implanted information
is removed while securing the picture substance’s privacy. This procedure is
helpful in applications, for example, restorative imaging where the first
picture must not get adjusted after the implanted information are separated.
Basically, RDH is acknowledged by the utilization of repetition in the first
picture. The two vital prerequisites for RDH systems: deformity to be low and
the embedding capacity to be huge.

 

 

CHAPTER 2

LITERATURE
SURVEY

 

2.1 A
reversible data hiding method for encrypted images.

W. Puech, M.
Chaumont and O. Strauss 1 first proposed in which the proprietor encodes the
first picture by Advanced Encryption Standard, and the information hider
installs one bit in each of the block containing pixels, implying that the
inserting rate is bits per pixel. On the recipient side, the information
extraction and picture reconstruction are acknowledged by investigating the
standard deviation amid decryption of the marked encrypted picture. This
technique requires that picture decoding must be done alongside information
extraction operations. At the end of the day, decoding and extraction are
indistinguishable.

 

2.2 Lossless
and Reversible Data Hiding in Encrypted Images with Public Key Cryptography.

Xinpeng Zhang,
Jing Long, Zichi Wang, and Hang Cheng 2 proposed a lossless and reversible
information concealing method for public key-encoded pictures with
probabilistic and homomorphic properties of cryptosystems. With such plans, the
pixel division and rearrangement is evaded and the encryption/decryption
operation is performed on the cover pixels specifically to such an extent that
the measure of scrambled information and the computational intricacy are
brought down. Because of information implanting on encoded space may bring
about less deformity in plaintext area due to the homomorphic property, the
inserted information can be separated trailed by the recuperation of actual
content from the unscrambled picture. With such joined strategy, the
beneficiary can remove some portion of implanted information earlier
unscrambling, and concentrate another piece of embedded information and recover
the actual plaintext picture after decoding.

 

 

2.3 Reversible
data hiding in encrypted images by reserving room before encryption.

K. Mama, W. Zhang,
X. Zhao, N. Yu, F. Li, 3 proposed a novel strategy for RDH in encoded
pictures, for such technique they don’t “vacate room after
encryption” and “reserve room before encryption” with
conventional RDH calculation, and along these lines it is very simple for the
information hider to install information in a reversible way into the scrambled
picture. The proposed technique can accomplish the reversibility that is
information extraction and picture recuperations are conceivable with no
mistake. As a matter of first importance they exhaust out the room with inserting
least significant bits of a few pixels into different pixels with a
conventional RDH strategy; afterward scrambled the picture, so that the places
of these least significant bits in the encoded picture may be utilized later to
shroud the information. The proposed technique not just accomplishes a
different information extraction from picture unscrambling yet additionally
accomplishes magnificent performance.

 

 

2.4 An improved reversible data hiding in encrypted
images using side match.

W. Hong 4 proposed a superior form of reversible
information concealing strategy in encoded pictures, which partitions the
scrambled picture into blocks, and each of the block conveys one bit by
flipping three least significant bits of an arrangement of pre-characterized
pixels. The information extraction and picture recuperation can be gotten by
checking the piece smoothness. Information recuperation of piece is performed
in plunging request of the outright smoothness distinction between two
competitor blocks. To additionally decrease the error rate the side match
procedure is conveyed.

 

 

2.5 Secret-fragment-visible mosaic image–a new
computer art and its application to Information hiding.

I. – J. Lai and W. – H. Tsai 5 proposed a picture
transformation technique, which chooses an objective picture like the secret
picture, at that point replaces each of the blocks in the objective picture by
a quite similar block of the secret picture and implants the map amongst secret
and target blocks; it shapes an encoded picture of the secret picture. To
locate the most similar block a greedy search strategy is utilized. This
strategy is reasonable for an objective picture comparative with the secret
picture, and the visual nature of encoded picture isn’t all that great in spite
of the fact that it is reversible.

 

 

2.6 A Lossless Data Hiding Method by Histogram
Shifting Based on an Adaptive Block Division Scheme.

Che-Wei Lee and Wen-Hsiang Tsai1 6 proposed lossless
information concealing strategy on the basis of histogram shifting technique.
This strategy utilizes a plan of adaptive division of cover pictures. The cover
pictures are separated into the blocks to yield expansive information
concealing limits. Picture qualities has been progressed. The technique is to
break a bottleneck of information concealing rate expanding at the picture. The
picture square size of 8×8 size. Such procedures are found in existing
histogram moving strategies. Four methods for block divisions are planned. The
one which gives the biggest information concealing limit is adaptively chosen.

 

 

2.7 High Capacity Lossless Data Embedding Technique
for Palette Images Based on Histogram Analysis.

N.
A. Saleh, H. N. Boghdad, S. I. Shaheen, A. M.Darwish 7 proposed, Data
hiding in encrypted image as one of the way of transmission of data securely
but due to technical aspects after the data extraction there may be large loss
in the image quality as well as original content of image. In the beginning
there are problem regarding the capacity of carrying data but recently this
problem is overcome. There is a possibility of huge loss in original quality of
image but it can carry high capacity data. This would mean that we are facing
only the problem of original content of image. To obtain solution for this problem,
the focusing area is Histogram. Histogram is nothing but graph of image which
is drawn on x-axis and y axis with references to the pixel values. Histogram analysis
indicates that we plot the graph of original image to avoid the loss of quality
of image after the issues with decryption.

 

 

 

 

 

 

 

 

 

 

 

CHAPTER 3

METHODOLOGY

 

 

 

3.1. Block Pairing

To influence the changed picture J’ to look like target picture J,
after change, each changed piece will be having the nearest mean and standard
deviation (SD) with the objective square. So we initially process the mean
value and standard deviation of each square of I and J individually. Give a
square B a possibility to be an arrangement of pixels with the end goal that B =
{p1, p2,. . . . pn}, and afterward the mean and standard deviation of this
square is computed as takes after:

                              

While coordinating pieces between unique picture and target
picture, we match obstructs with nearest standard deviations. To recuperate the
first picture from the changed picture that is transformed image, the places of
the first squares ought to be recorded and inserted into the changed picture
with a RDH technique. We initially group the squares as indicated by their standard
deviation esteems before blending them up. Actually, we found that the standard
deviation estimations of most squares pack in a little extend near zero and the
recurrence rapidly falls down, along with the expansion of the standard
deviation esteem. This way, we are able to separate the pieces into classes of two
with unequal extents: class 0 for hinders with littler standard deviations, and
class1 for obstructs with bigger standard deviations, and match up the squares
having a place with a similar class. By appointing the greater part of pieces
to the class 0 we can keep away from the vast deviation of standard deviations
between a couple of squares and proficiently pack the lists in the meantime. In
this paper, we propose to isolate both the first and target pictures into
non-covering 4 X 4 pieces and compute the standard deviations of each square.
We first gap the squares of unique picture I into 2 classes as indicated by the
quantile of standard deviations and relegate a class mark, 0 or 1, to each
piece. We filter the objective picture and mark first n 0 hinders with the
littlest standard deviations as Class 0, and the rest n 1 obstructs as Class 1.

 

 

A
basic case on the proposed piece blending technique is appeared in the Figure 1,
within which the picture just comprises of ten squares. By setting ? = 70, we
relegate 7 obstructs with littlest SDs into class 0, and the rest 3 hinders
into class 1 in the first picture. In the objective picture, in spite of the
fact that the eighth and ninth piece have a similar SD esteem 5, the eighth
square is appointed to class 0 yet the ninth square is allocated to the class
1, since class 0 can just incorporate 7 blocks as dictated by the class 0 of
the first picture. Subsequent to naming the class lists, we get a class index
table (CIT) for unique picture and target picture individually, which will be
useful for understanding the method of piece blending. As indicated by the
blending standard, the main square of the first picture is combined up with the
forward piece of the objective picture, in light of the fact that the two is the
principal square of class 1 as appeared in the class index table; the second
square of unique picture is matched up with the ninth piece of target picture,
on the grounds that two is the second square of class 1, etc. The blending
result is recorded in Figure 2, which can be created by the class index table
of unique picture and the class index table of the objective picture.

 

For
each combine of squares (B, T), the first piece B will be changed to target
piece T by mean moving and square turn, yielding T’. By supplanting every T
with T’ in the objective picture, the sender will produce the changed picture.
Note that the two operations of mean moving and square revolution won’t change
the standard deviation esteem, so T’ has an indistinguishable standard
deviation from B. In this way, the standard deviations in changed picture is
just a stage of those in unique picture. Along these lines, to re-establish the
first picture from the changed picture, the recipient just has to know the standard
deviation of the first
picture.

 

Fig.
1 Example of Block Pairing a) Original Image b) Target Image c) Transformed Image

 

Fig.
2 Block pairing result of the example Fig. 1

 

3.2. Block Transformation

By
the piece matching technique depicted above, in each combine (B, T), the two
squares have close standard deviation esteems. Hence, while changing B towards
T, we just need a mean moving change that is reversible.

 

 

Let
the first square B ={p1 , p2 , . . . , pn }, and the relating target piece T
={p’1 , p’2 , . . ., p’n }. With Eq. (1), we ascertain the methods for B and T
and mean them by uB and uT separately. The changed piece T’= { p”1 , p”2 , .
. . , p”n} is created by the mean moving as takes after:

where (uT ? uB ) is the distinction between the methods for target
square and unique piece. We need to move every pixel estimation of unique piece
by adequacy (uT ? uB) and in this way the changed square has a similar mean
with the relating target piece. Notwithstanding, in light of the fact that the
pixel esteem p”i ought to be a number, to keep the change reversible, we round
the distinction to be the nearest whole number as

            ?u = round(u T ? u B )           (4)

 

and move the pixel esteem by ?u, in particular, each p”i is
gotten by                             

                      

Note
that the pixel esteem p”i ought to be a whole number in the vicinity of 0 and
255, so the change (5) may bring about some finished stream/undercurrent pixel
esteems. To maintain a strategic distance from such changed squares went
without by (5), we accept that the most extreme flood pixel esteem is OVmax for
?u ? 0 or the base undercurrent pixel esteem is UNmin for ?u < 0. In the event that flood/undercurrent happens in a few squares, we kill them by changing ?u.        We utilize the adjusted ?u to move the pixels of piece B, and in this manner every one of the pixels' esteems are controlled into the scope of 0, 255. However the scope of ?u's esteem is still substantial, which can't be proficiently compacted. In this way we additionally change ?u as     in which the quantization step, ? is a much parameter. At that point it simply needs to record ?u' = 2|?u|/?, by which it has the benefit of not to record the indication of ?u. Since when ?u' is a much number it implies ?u ? 0 and when ?u is an odd number it implies ?u < 0. Since when ? is expansive the measure of data recording ?u' will be little however the counterbalance between the altered ?u and the first ?u will be huge, a tradeoff must made by picking ?. We set ? = 8 in the accompanying analyses. Finally, to keep up the comparability between the changed picture and target picture however much as could be normal, we furthermore transform the moved square into one of the four headings 0?, 90?, 180? or 270?.     The ideal heading is decided for limiting the root mean square error (MSE) between the pivoted piece and the objective piece. Subsequent to moving change and revolution, we get another piece T'. With these new squares, we supplant the comparing hinders in the objective picture and produce the changed picture J'. The parameters, ?u and turn headings, will be packed, scrambled and after that installed into the changed picture J' as AI to yield the "encoded picture" E(I) brought in this paper picture. The change and hostile to change methodology of the proposed technique are portrayed in Algorithm 1 and Algorithm 2 individually.             RIT produces an encoded picture E(I), which has the upside of keeping a significant type of the picture contrasted with conventional encryption strategies. Along these lines, it is free for the cloud server to utilize any established RDH on the scrambled picture. Choosing what sort of RDH technique relies upon whether to keep the picture quality or not. In this segment we just receive two RDH strategies, one is a conventional RDH that keeps the nature of pictures and the other is a brought together information inserting and scrambling technique that may incredibly debases picture structures for installing huge payload.                    CHAPTER 4 RESULTS AND DISCUSSION   In this segment trial comes about on the proposed RIT technique are introduced. 100 sets of pictures are haphazardly picked as our test pictures from the BossBase picture database.2 Firstly every one of the pictures are preprocessed to get a similar size of 1024 × 1024 pixels. Since in the RIT technique the parameter ? affects the AI payload, we give the test to choose a superior ? to enhance the general execution. The littler the space involved by AI is, the better the scrambled pictures visual quality will be. For ? in the scope of 0.05, 0.95, the variety of AI payload is by all accounts not vast. What's more, it can be seen that when ? is 0.75, the AI payload achieves the valley esteem. In the investigation, we list the "unscrambled pictures" with the correct key and the wrong key individually. The first picture can be losslessly reestablished with the right key.     Fig. 5. Experimental results of test images. (a) Original image. (b) Target image. (c) Encrypted image. (d) Decrypted image (right key). (e) Decrypted image (wrong key). The scrambled images E(I) got by RIT look like mosaic pictures with their appearance like the objective pictures. Since the distinction between the encoded picture and the objective picture is little, such visual impact will meet the necessity of disguise, which implies that the first picture content is completely secured by an objective picture content. Regardless of whether the aggressor perceives the disguise, without the mystery keyK of AES, it is additionally unfeasible to unscramble the AI that is essential for reestablishing the first picture. Furthermore, in this manner the assailant just gets an aimless picture as appeared in Fig. 5(e).   The quality of the encrypted image E(I) is measured by the peak-signal-to-noise ratio (PSNR) defined as PSNR =10 × log 10 ( 2552 /MSE ), where the MSE for m×n image is computed by formula in which xij and yij mean the pixel estimations of the objective picture J and encoded picture E(I), separately .It can been seen that AI possesses around 0.521 bits for every pixel (bpp) overall. Such vast overhead reason huge disfigurement to some degree, however the scrambled picture E(I) still can keep a satisfactory quality with the PSNR esteem about equivalent to 30 dB, which is an acknowledged visual impact. Other than we additionally give the normal payload of AI and PSNR estimation of 100 arbitrarily chose tried pictures, 0.529 bpp and 27.2 dB separately.                         CHAPTER 5 CONCLUSIONS   Unique in relation to past structures which scramble a plaintext picture into a ciphertext frame, reversible image transformation-based reversible data hiding –encrypted images shifts the semantic of unique picture to the semantic of another picture and in this manner ensure the security of the first picture. Since the encoded picture has the type of a plaintext picture, it will maintain a strategic distance from the documentation of the inquisitive cloud server and it is free for the cloud disjoin to pick any of reversible data hiding strategies for plaintext pictures to implant watermark. We understand a reversible image transformation based strategy by enhancing the picture change method to be reversible. By reversible image transformation, we can change the first picture to a subjective chose target picture with a similar size, and reestablish the first picture from the scrambled picture losslessly. Two RDH techniques including PEE-based reversible data hiding and UES are embraced to implant watermark in the encoded picture to fulfill distinctive needs on picture quality and installing limit. A few intriguing issues can be considered later on, including how to enhance the nature of the scrambled picture and how to stretch out thought of reversible image transformation to sound and video.                                       CHAPTER 6 REFERENCES   1 M. Chaumont, W. Puech and O. Strauss, "A reversible data hiding method for encrypted images" Proc. SPIE, Secure, Forensics, Steganography, Watermarking Multimedia Contents X, vol. 6819, p. 68191E, Feb. 2008. 2 F. Li, K. Ma, N. Yu, W. Zhang, X. Zhao, "Reversible data hiding in encrypted images by reserving room before encryption," IEEE Trans. on Information Forensics and Security, vol. 8, no. 3, pp. 553-562, Mar. 2013. 3 Hang Cheng, Jing Long, Xinpeng Zhang, and Zichi Wang "Lossless and Reversible Data Hiding in Encrypted Images with Public Key Cryptography,"IEEE Trans. on Circuits and Systems for Video Technology, 2015. 4 T. Chen, W. Hong, H. Wu , "An improved reversible data hiding in encrypted images using side match," IEEE Signal Processing Letters, vol. 19, no. 4, pp. 199 -202, Apr. 2012. 5 I.-J. Lai and W.-H. Tsai, "Secret-fragment-visible mosaic image–a new computer art and its application to Information hiding," IEEE Trans. Information Forensics and Security, vol. 6, no. 3, pp. 936–945, 2011. 6 Wen-Hsiang Tsail and Che-Wei Lee1 "A Lossless Data Hiding Method by Histogram Shifting Based on an Adaptive Block Division Scheme", River Publishers, 2010. 7 H. N. Boghdad, A. M. Darwish, N. A. Saleh, S. I. Shaheen, "High Capacity Lossless Data Embedding Technique for Palette Images Based on Histogram Analysis," Digital Signal Processing, 20, pp. 1629?1636, 2010.