This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2015.2514097, IEEE
Transactions on Industrial Informatics

Distributed Secret Sharing Approach with

Cheater Prevention based on QR Code
Pei-Yu Lin, Member, IEEE

 The sharing scheme also is incapable of preventing cheaters in

Abstract—QR barcodes are used extensively due to their its real-world application.
beneficial properties, including small tag, large data capacity, A reliable distributed secret storage system with the QR code
reliability, and high-speed scanning. However, the private data of can be used in significant applications, such as offering secret
the QR barcode lacks adequate security protection. In this article,
we design a secret QR sharing approach to protect the private QR
management and authorization in e-commerce. Based on our
data with a secure and reliable distributed system. The proposed observations, our aim was to design a distributed secret sharing
approach differs from related QR code schemes in that it uses the system based on the QR barcode, thereby allowing a secret to
QR characteristics to achieve secret sharing and can resist the be split into pieces and shared among individual QR-tag owners
print-and-scan operation. The secret can be split and conveyed to ensure the privacy of the QR data. The secret data can be
with QR tags in the distribution application, and the system can revealed when qualified QR-tag owners cooperate.
retrieve the lossless secret when authorized participants cooperate.
General browsers can read the original data from the marked QR
Recently, most QR-related research [5-11] has used the
tag via a barcode reader, and this helps reduce the security risk of traditional image hiding manner or the traditional
the secret. Based on our experiments, the new approach is feasible watermarking technique without utilizing the characteristics of
and provides content readability, cheater detectability, and an the QR barcode. The image hiding schemes treat the QR tag as
adjustable secret payload of the QR barcode. a secret image and then embed the QR image into the special
domain [5, 6] or the frequency domain [7, 8] of a cover image.
Index Terms—QR barcode, error correction capability, secret Hence, the secret payload of such schemes [5-8] is equal to the
sharing, cheater prevention
QR data. These schemes do not operate on the QR tag directly,
so they are incapable of allowing the practice of hiding/reading
I. INTRODUCTION the secret into/from the QR code directly.
Watermarking schemes [9-11] embed a watermark into the
C OMPARED with a one-dimensional (1D) barcode, the
two-dimensional (2D) QR barcode [1-3] can store a larger
data payload and possesses the capability of correcting errors.
frequency domain of a QR image with Discrete Wavelet
Transform (DWT), Discrete Cosine Transform (DCT), and
Discrete Fourier Transform (DFT) to protect the copyright of a
The barcode data easily can be decoded and retrieved via an QR image. The maximum payload of a watermark depends on
automatic barcode system. However, the lack of security of the the size of the QR image. The computational load of the
barcode with private data creates problems for its real-world transform operations of schemes [7-11] is more complex and
application. heavier than that of the schemes themselves [4-6, 12].
In general, to protect the privacy of the barcode data, the data Considering low-power barcode devices, the computational
normally are stored in a back-end database, and the barcode complexity of the QR code scheme should be minimized.
shows the web link for the database. Only a browser with the Avoiding the conventional schemes, Gao and Sun [12]
right access can log into the database and obtain the private data embedded the watermark into the QR tag by directly adjusting
[4]. However, the web link of the back-end database creates a the widths of the rows and columns of the QR module.
potential risk in that it may attract the intruder’s attention. Nevertheless, the embeddable payload of the watermark was
Chuang et al. [4] proposed a secret sharing scheme for the QR less than that of the related schemes [9-11] due to the limitation
tag in order to protect the secret barcode data. Unfortunately, of the adjustable widths of the rows and columns of the QR
the content of the QR tags is meaningless, and the shares easily modules. The scheme presented in [12] also was incapable of
can be obtained by scanning the QR tags with a barcode reader. extracting the correct watermark when the widths of the rows
and columns of the QR modules were distorted. The result was
Manuscript received December 04, 2014; revised October 02, 2015; that the scheme required an additional bilinear interpolation
accepted December 18, 2015. This work was supported by the National Science transform, morphological repair and BCH(15, 5) error
Council, Taiwan, under contract No. NSC 102-2221-E-155-035-MY3 and
MOST 104-3115-E-155-002.
correction to help extract the watermark from the distorted QR
Copyright © 2009 IEEE. Personal use of this material is permitted. However, tag.
permission to use this material for any other purposes must be obtained from The related schemes [4-11], however, do not consider and
the IEEE by sending a request to [email protected].
The author is with the Department of Information Communication, and
use the characteristics of the QR barcode. The error correction
Innovation Center for Big Data and Digital Convergence, Yuan Ze University, capability of the QR code allows the barcode reader to retrieve
Taiwan (email: [email protected]).

1551-3203 (c) 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2015.2514097, IEEE
Transactions on Industrial Informatics

the data correcttly if portionss of the barcoode are damagged. In is its reliability, which
w allows thhe barcode reaader to recoverr data
exxploring the errror correctionn capability of the QR techhnique, corrrectly even if pportions of thhe barcode are dirty or damaaged.
wee designed an efficient and feasible secrett QR sharing ssystem To achieve reliabbility, the QR code standardd offers four error
to protect the pprivate QR datta in real-worlld applicationns. The corrrection levels, i.e., L, M, Q, and H for eacch QR versionns, as
neew system moodifies the QR R modules direectly and can satisfy listeed in Table I. For insstance, level H can tolerate
the essentials of steganoggraphy, readaability, robusstness, approximately 300% of misdecoodes or substittution errors inn the
seecurity, low ccomputationall complexity and feasibiliity for dataa and error coorrection codew words. Here, the codewordd is a
distributed QR application. T To prevent dishonest particcipants unitt in the QR tagg that is equal to eight moduules.
froom obtaining the
t data, the deesigned sharinng scheme can verify
chheaters before revealing the shared secrett. The generateed QR TAB
BLE II
MAXXIMUM CHARACTE ER STORAGE CAPA ACITY
tagg can achievee more satisfactory robustneess than relateed QR
N
Number mber of Number of Number
of Num
scchemes [9-12]. Note that ouur scheme is noot limited to thhe QR Error
error errror data of data
Nuumber
coode, because itt can be applieed to the relateed 2D barcodees with Verssion correction off data
correction corrrection codeworrds codewor
level bl ocks per bloc
ck bbits
errror correctionn capability, such as the PDF417 andd Data c
codewords ds
M
Matrix codes. 1 L 7 1 19 19 152
The rest of the
t article is organized as follows. Secttion II M 10 1 16 16 128
introduces the QR
Q barcode teechnique. The proposed secrret QR 1
Q 13 13 13 104
shharing mechannism is presennted in Sectioon III. The sharing
H 17 1 9 9 72
simmulation andd performancee comparisonns are analyzzed in
Seection IV. Ourr conclusions are
a presented in Section V. 3 1077
200 L 224 861 66,888
5 1088
3 41
M 416 669 55,352
13 42
15 24
Q 600 485 33,880
5 25
15 15
H 700 385 33,080
10 16
19 1188
400 L 750 2,956 233,648
6 1199
18 47
M 1,372 2,334 188,672
31 48
34 24
Q 2,040 1,666 133,328
34 25
20 15
H 2,430 1,276 100,208
61 16

Figg. 1. Basic structuure of a QR barcoode.

T
Table II brieflyy presents the data payload and the reliabbility
TA
ABLE I of vvarious QR veersions and errror correctionn levels of thee QR
THE RELIABILITY
Y OF THE QR BARC CODE stanndard. Accordding to the QR version and error correcction
Error correction capabiility, % of codewoords leveel, the data coodewords in tthe QR tag aare segmentedd and
Errror correction levvel
(approox.)
storred into one oor more blockks. For instancce, the data inn QR
L (Low) 7%
verssion 1-L are 1152 bits (19 daata codewordss8 modules)) and
M (Medium) 15 %
are stored in one bblock. The datta in QR versioon 40-L are 233,648
Q (Quartile) 25 %
bits (2,956 data coodewords8 modules)
m and are segmentedd and
H (High) 30 % storred in 25 bloocks (19+6), i.e., 19 blockks each of which w
contains 118 datta codewords and six blocks each of which w
contains 119 daata codewordds. Then, thee error correcction
II. TECHNOLOG
GY OF QR BAR
RCODE coddewords that coorrespond to thhe data codewwords of each bblock
The QR barcoode (quick respponse code) iss an extensivelly used are generated to ensure the errror correctionn capability of the
2DD matrix repreesentation thaat was developped by the Jappanese blocck data.
Deenso-Wave C Company in 1994 [1-3]. A QR barcoode is O
Obviously, thee larger QR veersion and erroor correction llevel
coonstructed of square modules with white annd black squarre dots can offer higher data payloadd and reliability. To designn an
that represent thhe digits zeroo and one. Figg. 1 shows thee basic efficcient and feaasible applicaation for the QR barcode,, the
strructure of the QR tag, such as the positionn detection paatterns, propposed schemee exploits thee adjustable capacity
c and error
aliignment patteerns, format information, version
v inform
mation, corrrection featuree to achieve reeadability andd secret sharinng on
annd the data annd correction ccodewords. T The QR modulles are QR modules direcctly.
suurrounded by a blank, quiet--zone border.
The QR codee standard [1] provides 40 Q QR versions too carry
vaarious data payyloads. The laarger QR version can offer higher III. SECRET (N, N)-THRESHOLD
D QR CODE SH
HARING APPRO
OACH
daata payload. AAnother significcant property of the QR techhnique B
Based on the prroperties of thee QR techniquue, the new schheme

1551-3203 (c) 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2015.2514097, IEEE
Transactions on Industrial Informatics

deesigns an (n, n))-threshold shaaring system sso that the privvacy of accoording to the QQR version annd the number of blocks b:
a secret
s is providded, making itt unavailable tto a cheater. Inn the (n, lv  (1  α  QR version)  b , (2)
n))-threshold shaaring system, a dealer and n participantss exist, wheere the parameeter α is the strength
s of cheeater detectabbility.
whhere n ≥ 2. Thhe dealer is reesponsible forr splitting the secret Thee value of α is a real numbber that can bbe adjusted byy the
into n marked QRQ tags, as outtlined in Sectioon III-A. Afterrwards, (C / b )  1
the n marked QR R tags can be distributed to the n correspoonding dealler, where 0  α  .
QR versiion
paarticipants. Onnly the n partiicipants with authorized QR R tags
arre qualified to obtain the shhared secret, annd no subset oof less Stepp 3: Estimate the length of ssecret S, ls, byy the equation:
than n tags can lleak any inform
rmation about the secret (as shown ls = C－llv. (3)
in Section III-B). With the preliminary phase, the payloaads of secret and
W
The proposedd secret sharingg with QR codde can be appliied for authhentication codde can be learnned accordingg to the versionn and
vaalue-added barrcode applicatiions, such as tthe distributedd secret the error correctiion level of thhe given QR barcode. Thiss can
shharing, e-coupoon, and e-tickeet. Fig. 2 show
ws the architectture of guarantee that thhe new algorrithm can lim mit distortion and
the proposed syystem. presserve the readaability of the QQR content.
F
For instance, giiven a QR barccode with verssion 20-L, as llisted
in T
Table II, there are eight bloccks in the 20-L L QR barcode (i.e.,
b = 8) and the num mber of error ccorrection coddewords, E, is 224.
Hennce, the modifiable capacity, C, equals to
C  224 / 2  8  896 moduules. The leength of V is
lv  (1  0.5  200)  8  88 bitss, where α = 0.5. Accordinng to
the values of C aand lv, the capaacity of S can be derived ass ls =
8966−88 = 808 bitts.
Shaadow Derivatioon Phase
T
To derive the ssecret shadow ws from S and V, first, the deealer
assiigns n secret kkeys, ki, for thee n correspondding participannts, i
= 1, 2, …, n. Thhe process of shadow generration is descrribed
beloow:
Step p 1: Derive tthe master keey K by the pparticipants’ secret
keys:
n
K  ki . (4)
i 1
Step p 2: Generate n authenticatiion streams, vi, and the lenggth of
Figg. 2. Architecturee of the proposed system.
vi iss lv:
vi = HK(ki), i = 1, 2, …, n. (5)
A. Secret Sharing Proceduree Herre, HK(·) is a onne-way hash function
fu [13] w
with the masterr key,
Assume that QRi are the n covers of QR R barcodes with the K.
saame QR versioon and error coorrection levell, i = 1, 2, …, n, and Step p 3: Generate (n−1) random m binary shadoows, s1, s2, …, sn−1,
the data of QRi can be different. That is, thee barcode readder can withh each having the length of ls.
sccan and decodee diverse data from QRi. Lett S be the privaate QR Step p 4: Derive thhe n-th binary shadow, sn, wwith length ls byy the
daata to be proteccted. shaddows s1, s2, …
…, sn−1 and the secret S:
sn = s1⊕s2⊕…⊕ ⊕sn−1 ⊕ S, (6)
Prreliminary Ph hase
wheere ⊕ denotess the bit stream m exclusive-orr (XOR) operaation.
According too the QR codee core architeccture of QRi, let the
nuumber of bloccks be b andd the number of error corrrection Herreafter, the deaaler can learn n shadows si and authenticaation
coodewords be E.E To share the secret S with cheater detecttability, streams vi, i = 1, 2,
2 …, n.
ann authentication code V iis used to verify v the invvolved Con ncealment Phaase
paarticipants. Baased on the obsservation of thhe QR algorithhm, the W
With exploringg the QR codee core algorithm m, the new cooding
errror correctionn capability iss less than hallf of the num mber of scheeme conceals the (si, vi) within
w the moodules of the data
errror correctionn codewords. Accordinglyy, the new scheme coddewords of QR Ri only, leavinng all other modules
m of thee QR
deetermines the ppayloads of S and
a V dynamiccally, along w with the codde unchanged.
QR R version andd the error corrrection level, aas follows: F
For the given QRi, let m bbe the modulees of the QR data
Sttep 1: Calculaate the value oof modifiable capacity, C, ffor the coddewords and lm be the lengtth of m. Here,, lm equals thee QR
given QR barcoode as: dataa codewords multiplied byy eight moduules. (A codew word
C  E 2   8 (1) refeers to eight moodules [1].) Forr instance, therre are 861 QRR data
whhere E is the nnumber of erroor correction codewords.
c coddewords in QR R version 20-L L. Hence, lm = 8618 = 6,888 6
Sttep 2: Evaluaate the length of the authenntication codee V, lv, moddules.

1551-3203 (c) 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2015.2514097, IEEE
Transactions on Industrial Informatics

Afterward, the (si, vi) are concealed into portions of the m data help reduce the suspicion of general QR users while they
modules of QRi, i = 1, 2, …, n, by applying the wet paper codes are scanning the QR code. Moreover, browsers and the
(WPCs) algorithm [14]. The process is described as follows: involved participants are incapable of decoding and extracting
Step 1: Treat the m modules of the data codewords in QRi as the shared secret without sufficient QRi' and keys.
matrix Mi with the size of lm1.
B. Secret Revealing Procedure
Step 2: Select the number of ls modules in Mi randomly as dry
elements, and the remaining number of (lm−ls) modules are In real-world applications, the capability of detecting
treated as wet elements, i = 1, 2, …, n. Note that the number of cheaters is a significant requirement before the secret data are
ls modules is selected evenly from the b blocks to preserve the revealed. In the proposed (n, n)-threshold sharing approach,
error correction capability of each block of QRi, i.e., ls/b only sufficient participants with the n-validated QR tags and
modules are picked up from each block. For instance, if we let ls keys can cooperate to reveal the shared secret. The designed
= 808 and lm = 6,888, we thereby select 808 modules randomly approach can detect dishonest participants and identify who the
as dry elements from the 6,888 QR data modules. cheaters are. In addition, the revealing procedure of the
Step 3: Generate the binary matrix Di with size lslm by the proposed scheme is blind, i.e., the authorized participants can
key ki, i = 1, 2, …, n. extract the secret without the host QR barcode and additional
[ Di ]lslm  RNG (k i ) , information.
(7)
where RNG(ki) is a random number generator using the key ki Assume that QR i and ki are the n-provided QR barcodes
as the initial seed. and keys, respectively, from the involved participants, i = 1,
Step 4: Adjust the ls dry elements from Mi to Mi' by complying 2, …, n. By utilizing a barcode reader, the information of the
with the following formula: QR version, the error correction level and the related formats
[ Di ]lslm  [ M i ]lm1  [ si ]ls1 . (8) can be recognized from QR i immediately. Let E be the
Here, the matrix [si] is formed by the shadow si with size ls1, i number of error correction codewords, and let b be the number
= 1, 2, …, n. The modified result [Mi'] can be derived by of blocks. The detection of cheaters and the extraction of the
rewriting or retaining the ls dry elements in [Mi] based on the secret S can be performed by the following phases:
solvability of the linear equations.
Estimation Phase
Step 5: Regard the ls modules of Mi' in Step 4 as wet elements,
and then select lv modules randomly from the remaining (lm−ls) Step 1: Calculate the values of C and m according to the QR
modules in Mi' as dry elements. Subsequently, the remaining barcode core by the definitions, C  E 2  8 and m = QR
modules (lm−ls−lv) are regarded as wet elements, i = 1, 2, …, n. data codewords8.
Note that the lv modules also are selected evenly from the b Step 2: Estimate the value of lv according to the QR version, the
blocks, i.e, lv/b modules are picked up from each block. number of blocks, and the pre-shared parameter α:
Step 6: Generate binary matrix Di' with size lvlm by the lv  (1  α  QR version)  b . (11)
master key K and ki, i = 1, 2, …, n: Step 3: Evaluate the value of ls by
[ Di ]lvlm  RNG K (k i ) , (9) ls = C－lv. (12)
where RNGK(ki) is a random number generator with the master n
key K and using the key ki as the initial seed. Step 4: Generate a master key K by K   ki .
Step 7: Modify the lv dry elements in Mi' with symbols Mi'' to i 1
comply with the formula, Step 5: Reproduce the n authentication streams, vi , each with
[ Di ]lvlm  [ M i]lm1  [vi ]lv1 . (10) length lv by the formula:
Here, the matrix [vi] is the authentication stream vi with size
vi  H K (k i ), i  1, 2, ..., n , (13)
lv1. [Mi'] can be rewritten to the modified result [Mi''] based
on the solvability of the linear equations. where H K () is a one-way hash function with the master key
Step 8: Learn the marked QR code, QRi', by replacing the m K.
modules in the data codewords of QRi with the m elements of
Cheater Identification Phase
the results Mi'', i = 1, 2, …, n.
Subsequently, the marked QR code, QRi', along with the key, ki, Step 1: Generate n binary matrixes, Di , with size lvlm by k i ,
can be shared and distributed to the involved i-th participants, i i = 1, 2, …, n.
= 1, 2, …, n.
[ Di ]lvlm  RNGK (ki ) , (14)
As we expected, the marked results, Mi', of the first level
concealment in Step 4 can guarantee that, at most, only ls where RNGK (ki ) is a random number generator with the
modules could be changed. Furthermore, the second level
master key K, and k i is used as the initial seed.
concealment in Step 7 also ensures that, at most, lv modules
could be altered in Mi''. That is, the designed algorithm can Step 2: Regard the m modules in the QR data codewords of
limit the distortion within C modules of the given QR code. QR i as a matrix M i with size lm1, i = 1, 2, …, n.
This leads to the barcode reader’s being able to scan and decode
Step 3: Obtain the authentication results, vi , according to the
the data from Mi'' successfully. The extracted meaningful QR

1551-3203 (c) 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2015.2514097, IEEE
Transactions on Industrial Informatics

foormula: marrked QR code from the geneeral QR users. However, genneral

[vi]  [ Di ]lvlm  [M i ]lm
l 1 , i = 1, 2, …
…, n. (15) browwsers and thhe involved participants are incapablee of
extrracting the shaared secret withhout sufficientt and validatedd QR
ded QR i and ki by
Sttep 4: Verify tthe genuinenesss of the provid barccodes with keyys.
coomparing vi w
with vi , i = 1, 2, …, n. If vi differs froom vi ,
the QR i is indiicated as “tam mpered” for thee i-th participannt, and
the secret revealling proceduree will be termiinated. Otherw wise, if
vi is equal to vi , i = 1, 2, …,, n, the providded QR barcoddes and (a) Cover QR code 1 (b) Cover QR code 2 (c) Cover QR coode 3
keeys are regardeed as “validateed,” and the seecret retrievall phase
caan be performeed in the next process.
Seecret Retrieval Phase
(d) M
Marked QR code of (e) Markedd QR code of ((f) Marked QR coode of
Sttep 1: Construuct n binary maatrixes, Di , w
with size lslm by the (a) ((b) (c)

keey ki , [ Di]lslm
m  RNG ( k i ) , i = 1, 2, …, n n, where RNG((ki) is a Fig. 3. The (3, 3)-thhreshold sharing oof the proposed m
method for QR veersion
1-L.
raandom numberr generator usiing the key ki aas the initial seed.
Sttep 2: Derivee n shadow m matrixes, [ si ] , i = 1, 2, …, n, by F
Figs. 4(a)-(c) show the oriiginal 40-H QR Q barcodes with
peerforming the bbinary multipllication as diffferent data (upp to 10,208 biits). Accordinng to the propposed
[si ]  [Di]lslm [Mi ]lm1 . (16) secrret sharing prrocedure, the new approachh can share 9,315 9
secrret bits and 4005 authenticattion bits (heree, α = 0.1) intoo the
Sttep 3: Computte the secret m matrix, S , withh the length off ls1, QR codes and foorm the corressponding marrked QR codees, as
S = s1 ⊕ s2 ⊕…⊕ sn . (17) show wn in Figs. 4(d)-(f). Considering the readability off the
Thhe authenticaated participan nts eventuallyy can disclosse the marrked QR dataa, the designned algorithm m can satisfyy the
orriginal secret S by regarding the two-dimennsional matrixx S to esseentials of steg
ganography annd readability
y by exploringg the
onne-dimensionaal bit stream. char racteristics of
f the error corr ection capabillity of the QR tag.

IV. SIMULATION RESULTS AND ANALYSIS

To evaluate the applicabillity of the prooposed schem me, the
oppen source librrary, ZXing libbrary [15], witth C#.NET lan nguage
waas used to gennerate and deccode the multii-format barco ode for
crross-platform ddevelopment. T The smallest vversion, 1-L (vversion (a) Cover QR code 1 (b) Cover QR code 2 (c) Cover QR coode 3
1, error correction level L), and the larrgest version, 40-H
(vversion 40, erroor correction level
l H), are sselected as thee cover
QR R barcodes. TThe size of thee generated QR R barcode by Z ZXing
is set to 2502550 pixels.
A. Practicabiliity of the Secreet QR Sharingg
Fig. 3 and Figg. 4 show the results of the new scheme ffor the (d) M
Marked QR code of (e) Markedd QR code of ((f) Marked QR coode of
1--L and 40-H QR versions in the (3, 3)-threshold
3 sharing (a) ((b) (c)
syystem. Note thhat the designeed approach can be applied to the Fig. 4. The (3, 3)-thhreshold sharing oof the proposed m
method for QR veersion
H.
40-H
(nn, n)-thresholdd sharing systeem, while 2 ≤ n. Based on thhe QR
coore algorithm, the maximum m payload of tthe data for thhe 1-L
T
The detection and identificaation of cheateers are essentiial to
veersion is 152 bits. Figs. 3((a)-(c) show tthe normal 1--L QR
prevvent fraudulennt participantss before reveaaling the secrret in
baarcodes with th he data “Yuann Ze”, “YZU IC” and “info.ccom.”,
the sharing systeem. Accordinng to the cheeater identificaation
respectively. WWith α = 0.1, tthe new approoach can concceal 23
phase in Section IIII-B, the involved participannts can verify both
seecret bits and oone authenticaation bit into tthe 1-L QR baarcode.
the provided keys and the marrked QR codees with each oother.
Thhe correspondiing marked QR R codes are lissted in Figs. 3((d)-(f).
Witthout the genuuine keys, the derived
d authenntication codees are
Thhe marked QR R barcodes, allong with the keys, can be shared
disttinct from thee original onees at Step 4 aand Step 5 inn the
annd distributed tto the individuual participantts.
estimmation phase. In the cheaterr identificationn phase, Step 4 can
Because the QRQ barcode consists of noise-like symbools, the
valiidate the provvided QR coddes without reevealing the secret
modification off the marked Q
m QR barcode cannnot be discernned by
shaddows. Once thhe provided keeys and QR taags are regardeed as
the human eyess. With a barccode reader, geeneral browseers can
“vallidated”, the authenticatedd participantss are allowedd to
sccan and read th he QR data froom the markeed QR code. T That is,
perfform the secreet retrieval phaase.
geeneral browserrs can read the content “Yuaan Ze” from Figg. 3(d).
Thhe meaningfu ul content heelp reduce the suspicion oof the

1551-3203 (c) 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2015.2514097, IEEE
Transactions on Industrial Informatics

TABLE III
PAYLOADS OF AUTHENTICATION CODE AND SECRET OF THE MARKED QR BARCODE FOR DIFFERENT α VALUES

Error correction levels

Version α L (7%) M (15%) Q (25%) H (30%)
Secret Secret Secret Secret
Authentication (bit) Authentication (bit) Authentication (bit) Authentication (bit)
(bit) (bit) (bit) (bit)
1 0.1 1 23 1 39 1 47 1 63
1 2 22 2 38 2 46 2 62
5 0.1 1 103 2 190 4 284 4 348
1 6 98 12 180 24 264 24 328
10 0.1 8 280 10 510 16 752 16 880
1 44 244 55 465 88 680 88 808
15 0.1 12 516 20 940 24 1,416 36 1,692
1 96 432 160 800 192 1,248 288 1,440
20 0.1 24 872 48 1,616 60 2,340 75 2,725
1 168 728 336 1,328 420 1,980 525 2,275
25 0.1 36 1,212 63 2,289 87 3,393 105 4,095
1.0 312 936 546 1,806 754 2,726 910 3,290
30 0.1 60 1,740 116 3,132 160 4,640 192 5,568
1 465 1,335 899 2,349 1,240 3,560 1,488 4,272
35 0.1 76 2,204 152 4,104 212 6,148 252 7,308
1 684 1,596 1,368 2,888 1,908 4,452 2,268 5,292
40 0.1 125 2,875 245 5,243 340 7,820 405 9,315
1 1,025 1,975 2,009 3,479 2,788 5,372 3,321 6,399

In order to demonstrate the secret capacity and cheater dpi and then scanned with 200 dpi to obtain the digital images
detectability, Table III lists the amounts of the shared secret and without any correction or restoration. The QR data of the three
the authentication code for different values of α. The printed and scanned QR tags are readable by a barcode reader.
modifiable capacity, C, is determined by the number of error In addition, the QR tags with the validated keys can reveal the
correction codewords in (1). That is, the higher settings of the shared secret correctly. The designed technique is tolerant of
QR version and error correction level provide larger capacity the common print/scan operations to preserve the QR data and
for the secret. To cope with the shared secret payload and the secret shares in real-world application.
detectability, the designed scheme can select an adequate QR
version and error correction level dynamically to derive the TABLE IV
MARKED 1-L QR BARCODES AFTER PRINTING AND SCANNING
marked QR codes. The maximum secret payload could be 24 to
Marked QR code 1 Marked QR code 2 Marked QR code
9,720 bits for QR versions 1-L to 40-H, while α = 0.
Considering cheater detection, the value of α can be
increased to enhance detectability. In the designed system, the Print and
scheme offers ( 1  α  QR version ) authentication bits for scan
each QR block in (2). That is, the new scheme can validate the
genuineness of the marked QR code by justifying each QR QR content Readable
block. As the definition in (2), there is at least one Secret Decodable
authentication bit per block for the lower QR version. For the
higher QR version, i.e., 40-H, there are 81 QR blocks, and each Table V illustrates the ability of the marked QR code if the
block contains 41 authentication bits, while α = 1. marked QR codes suffered from common processes, such as
Considering the common situations of noise, rotation, and noise, blur, compression, and rotation. Noise occurs when
print of the QR code in real-world applications, a QR tag should communicating or capturing the QR tag in the real world. Here,
able to withstand various distortions. Statistically, the new the Gaussian and Uniform noises are added to the marked QR
scheme modifies C/2 modules of the original QR barcode by tags at 10%, 30%, 50%, and 70%. Moreover, the compression
changing or unchanging the dry modules. That is, the derived technique usually is used to compress the QR tag for reducing
marked QR tag can possess the readability of QR data and, the storage space. To evaluate the feasibility of the designed
theoretically, also maintain a portion of the error correction scheme under image compression, the JPEG 2000 lossy
capability. compression is mounted to the marked QR image with quality
The print and scan operation for the QR tag is practiced factors (Q) of 100% and 0%. The rotation is mounted further to
extensively in the commercial magazine, poster, and the marked QR tag to estimate the performance of the new
publication fields. Table IV lists the printed and scanned tags scheme with the use of mobile devices and various degrees of
for the marked 1-L QR codes (Figs. 3(d)-(f)). The marked QR scan.
tags were printed by an HP LaserJet M2727 printer with 600

1551-3203 (c) 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2015.2514097, IEEE
Transactions on Industrial Informatics

TABLE V
RESULTS OF THE MARKED 1-L QR BARCODEES AFTER COMMON
N PROCESSES
Maarked QR code
Gaussian
n Noise The original QR
R code, 70%
10% 30% 50% 7
70%

Distortedd QR code

QR conteent Readable Readable Readable Reaadable Readabble

Secret Decodable Decodable D
Decodable Decoodable −

Maarked QR code
Uniform Noise The original QR
R code, 70%
10% 30% 50% 7
70%

Distortedd QR code

QR conteent Readable Readable Readable Reaadable Readabble

Secret Decodable Decodable D
Decodable Decoodable −

Maarked QR code The original Q

QR code,
n blurring
Gaussian
Radius: 1 pixeels Radius: 2 pixels Radiius: 2.5 pixels Radiuss: 3 pixels Radius: 3 pixels

Distortedd QR code

QR conteent Readable Readable Readable Reaadable Readabble

Secret Decodable Decodable D
Decodable Decoodable −

Maarked QR code The original Q

QR code,
Compresssion
JPG, Q=100%
% JPG, Q=0% JPEG 2000 Q=100% JPEG 20000, Q=0% JPEG 2000,, Q=0%

Distortedd QR code

QR conteent Readable Readable Readable Reaadable Readabble

Secret Decodable Decodable D
Decodable Decoodable −

Maarked QR code The original Q

QR code,
n
Rotation
45∘ 90∘ 135∘ 2270∘ 270∘

Distortedd QR code

QR conteent Readable Readable Readable Reaadable Readabble

Secret Decodable Decodable D
Decodable Decoodable −

1551-3203 (c) 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2015.2514097, IEEE
Transactions on Industrial Informatics

TABLE VI
DECODING RESULTS WITH MO
OBILE DEVICES
Deevices iPhone 4s HTCC One Max Samsun
ng Note5

Gaaussian Noise,
70%

TABLE VII
COMPARISON O
OF RELATED QR BARCODE
B SCHEME
ES
Thhe proposed
Functionality [4] [5]] [6] [7, 8] [9-111] [122]
scheme
Application fieldd Seccret sharing Im
mage hiding Im
mage hiding Imaage hiding Wattermarking Waatermarking Seecret sharing
Domain Spaatial Sppatial Spatial Freequency Freqquency Spaatial Sppatial
Meaningful of thhe marked
Noo Yees Yees Yess Yess Yees Yees
result
Operation on thee QR code Noo Noo Noo No Yess Yees Yees
Module-based Noo Noo Noo No No Yees Yees
Utilizing the erroor
Noo Noo Noo No No Noo Yees
correction capability
Computational complexity Low
w Loow Loow Higgh Higgh Low
w Loow
Robustness Higgh Hiigh Loow Midd Midd Low
w Hiigh
Addjustable,
Secret capacity QR
R image QR
R image QR
R image QR
R image Midd Low
w
244~9,720 bits

To demonstraate the decodaability of the ddistorted QR tags t in cover image. Connsequently, thee secret capaccity of the scheemes
Taable V, we useed three differeent mobile devvices, i.e., an iiPhone [5-88] is equal to thhe data of the QR image.
4ss, HTC One M MAX, and Sam msung Note5. T The term “readdable” T
The watermarkking scheme [12] [ adjusts thhe widths of rrows
in Table V indiicates that the QR data of tthe distorted Q QR tag andd columns of the QR moddule to conceaal the waterm mark.
caan be decodeed correctly bby a barcodee reader. Thee term Hennce, the embedddable capaciity and the robbustness of [12] is
“ddecodable” dennotes that the distorted QR R tags with vallidated lesss than that off other schem mes [9-11]. TThe computatiional
keeys can reveall the shared seecret afterwarrd. It is obviouus that commplexity of scchemes [7-11]] with the frequency domaain is
the designed steeganography Q QR sharing syystem is practticable relaatively greaterr than that of the schemes [4-6, 12] withh the
annd has sufficiennt ability to resist noise, blurr, lossy comprression, spattial domain annd the proposeed system. Beecause the seleected
annd various scann angles. Tablee VI shows thee decoding ressults of coeffficients of tthe QR frequuency domainn are limited,, the
ouur marked QR R codes by thhe three mobille devices. Thhe QR capaacities of [7-11] are restricteed.
coontent “Yuan Ze” can be deecoded successsfully via thee three T
To comply witth various seccret capacities, the shared secret
different mobilee devices. payload of the prooposed approaach is adjustabble (24 - 9,720 bits)
For the sakee of comparinng the decodaability betweeen the alonng with an adeequate QR verrsion and erroor correction level.
orriginal QR codde with the dessigned QR codde, the last coluumn of Diffferent from rellated schemes, the designed approach expllores
Taable V providdes the origiinal QR tag (without the secret the characteristiics of a Q QR barcode to provide the
emmbedded) wheen subjected too distortion. Noote that the diffferent stegganography, readability, robbustness, andd adjustable secret
baarcode decodeers and enviironments [166], such as lights, capaacity for the seecret sharing mechanism.
m Thhe new mechannism
m
monitors, scalees, and mobile devices, could c influencce the conceals the secreet shares on thhe QR module directly. Therreby,
deecoded results of the originaal barcode and our marked QR Q tag. the modified secrret module off the designed approach is more m
robuust against geoometric attackks (as shown inn Table V) thann the
B. Comparison n and Discussiion
scheemes that havee modified LS SBs and coeffiicients [5-12].
Because articcles concerninng secret sharring on QR baarcode C
Chuang et al. [4] achievedd the distributted secret shaaring
arre relatively raare [4], to exppress the funcctionality of thhe QR application with a QR tag by using Shamirr’s (t, n)-thresshold
appplication, Tabble VII providdes an overalll comparison of the systtem. Their sccheme has thhe same robuustness and secret
QR R related dataa hiding schemmes, watermarkking schemes, secret capaacity as the QQR technique. However, thee derived QR tags
shharing schemees [4], and thhe proposed approach. Thhe QR are meaningless aand incapable of satisfying the steganography
hiding schemess [5-8] can bbe regarded aas the convenntional andd readability reequirements off the QR conteent.
im
mage hiding tecchnique that eembeds the seccret QR imagee into a

1551-3203 (c) 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2015.2514097, IEEE
Transactions on Industrial Informatics

C.. Security Analysis we pplan to investigate the Reedd-Solomon (RRS) code of thee QR
Considering the cheating situation, a ddishonest partiicipant techhnique to reduuce the modifications requirred and to impprove
m
may offer the ggenuine key annd a fake QR tag to cheat others.
o the security of thee QR barcode..
Leet I be the n authorized participants thaat can cooperrate to
reveal the secrret S in the (n, n)-threshhold secret sharing REFER
RENCES
mechanism. Thhe probabilityy that particippants of I   I will
m [1] ISO/IEC 18004,, “Information tecchnology – Autom matic identificatioon and
m data capture techhniques – Bar codde symbology – Q QR Code,” 2000.
 
[2] Denso-wave, [O Online], Availablee: http://www.qrcoode.com/en/indexx.html
, where I   n . The nootation
l
coorrectly reconsstruct S is 2  s [3] Psytec QR code editor software, [
[Online]. Avaiilable:
http://www.psyttec.co.jp/docomo..html
n
  is defined as the numbber of combinnations of n ddistinct [4] J. C. Chuang, YY. C. Hu and H. J. Ko, “A novel seecret sharing techhnique
k  using QR code,” International Journal of Imagge Processing, V
pp.468-475, 20110.
Vol. 4,

things taken k att a time. This guarantees thaat the participaants of [5] H. C. Huang, F. C. Chang and W. W C. Fang, ”Reveersible data hidingg with
I' are hardly abble to restoree the secret. It is difficultt for a histogram-basedd difference expannsion for QR Codde applications,” IEEE
fraaudulent partiicipant to passs the cheater identification phase Transactions onn Consumer Elecctronics, Vol. 577, No. 2, pp. 7799-787,
2011.
wiith a fake keyy or a tampereed QR tag. Thhe probability that a [6] S. Dey, K. Moondal, J. Nath annd A. Nath, “Addvanced steganoggraphy
chheater could suuccessfully hiit all of the auuthentication bits is algorithm usingg randomized inteermediate QR hoost embedded witth any
m encrypted secrett message: ASA_Q QR algorithm,” Innternational Jourrnal of
 
l Modern Educatiion and Computerr Science, Vol. 6, pp. 59-67, 2012..
2  v, which iss negligible. [7] C. H. Chung, W W. Y. Chen and C.. M. Tu, “Image hidden
h technique using
For instance, the probabilitty of hitting thhe secret withh fewer QR-Barcode,” F Fifth Internationall Conference on Intelligent Informmation
 152  Hiding and Multimedia Signal Prrocessing, 2009.
  
22  [8] W. Y. Chen and J. W. Wang, “Neested image stegannography scheme using
than n participaants for the 1-L QR versionn is 2   , aand the QR-barcode tecchnique,” Opticaal Engineering, Vol. 48, No. 55, pp.
 10 , 208  057004-01~057004-10, 2009.
  
6, 399  [9] M. Sun , J. Si and S. Zhang, “R Research on embbedding and extraacting
prrobability for tthe 40-H QR vversion is 2 
where α = 1. ,
methods for diigital watermarkss applied to QR R code images,”” New
Inn addition, an intruder is haardly able to iidentify the geenuine Zealand Journal of Agricultural R Research, Vol. 500, pp. 861-867, 20007.
m [10] L. Li, R. L. Wanng and C. C. Chaang, “A digital w watermark algorithhm for
 
l QR code,” Interrnational Journall of Intelligent Innformation Proceessing,
seecret informattion from thee 2  s
exhauustive combinnations
Vol. 2, No. 2, ppp.29-36, 2011.
wiithout the knoowledge of n-marked QR tags and keys. The [11] S. Rungraungsillp, M. Ketcham, V V. Kosolvijak and S. Vongpradhip, “Data
prrobability that a dishonest participant can fake a valid Q
QR tag hiding method ffor QR code baseed on watermark by compare DCT T with
 152  DFT domain,”” Internationall Conference on Computer and
  
2  Communication Technologies, ppp. 144-148, 2012.
to deceive otheers for the 1-L
L QR version is 2   , annd the [12] M. Gao and B. Sun,
S “Blind waterrmark algorithm based
b on QR barccode,”
 10, 208  Foundations off Intelligent Syystems : Proceeedings of the Sixth
  
3, 321  International Conference
C on Intelligent Systeems and Know wledge
prrobability for tthe 40-H QR vversion is 2 
, where α = 1.
Engineering, Vool. 122, pp. 457-462, 2011.
Too improve thee ability to dettect cheaters, the
t value of thhe real [13] H. Martin, P. Peeris-Lopez, J.E. TTapiador, E. San Millan, “An Estiimator
nuumber α can be increased further accorrding to the ssystem for the ASIC Foootprint Area of Lightweight
L Crypttographic Algoritthms,”
(C / b)  1 IEEE Transactiions on Industriial Informatics, Vol. 10, No. 22, pp.
deemand, wheree 0    . Thhis shows thhat the 1216-1225, 20144.
QRR version [14] Y. J. Chiang, P. Y. Lin, R. Z. Wang and Y. H. Chen, ““Blind
higher settings oof the QR verrsion, error corrrection level,, and α steganographic approach
a for QR code
c module baseed upon error correection
capability,” KSII
II Transactions onn Internet and Info
formation Systemss, Vol.
caan offer larger number of auuthentication bits to defend aagainst 7, No. 10, pp. 25527-2543, 2013.
a cheater. Onlyy the authoriized participaants can reveaal and [15] ZXing ("Z
Zebra Crossing"), [Onnline], Avaiilable:
deecrypt the secrret stream. http://code.googgle.com/p/zxing/
[16] D. Munoz-Mejjias, I. Gonzaleez-Diaz and F. Diaz-de-Mariaa, “A
low-complexity pre-processing syystem for restorinng low-quality QR R code
images,” IEEE T Transactions on Consumer
C Electronics, Vol. 57, No. 3, pp.
V. CONCLUSIONS AND FUTUR
RE WORK 1320-13, 2011.
Different froom the connventional QR R applicationn, the
prroposed approaach utilizes thee characteristics of QR moduules to Pei-Yu Lin (M’10) received the M.S. and Ph.D.
saatisfy the essenntials of stegannography, reaadability, robusstness, degrees fromm National Chunng Cheng Univeersity,
addjustable secreet capacity, bllind extractionn, cheater deteection, Chiayi, Taiwaan, in 2004 and 2009, respectivelyy, both
in computer sccience and inform
mation engineeringg.
annd identificatiion for the secret sharinng mechanism m. As She is currrently an Associaate Professor witth the
deemonstrated inn this experim ment, the new QR sharing ssystem Department off Information Coommunication at Yuan
caan achieve satisfactory perfoormance when compared to rrelated Ze University,, Taiwan. Her cuurrent research intterests
include image protection, data m
mining and informmation
atttempts. Also, the designed algorithm is feasible and can c be
security.
appplied to the related 2D barcodes wiith error corrrection
caapability, suchh as the PDF4117 and Data M Matrix. In the ffuture,

1551-3203 (c) 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.