2022

2022 IEEE
2022 IEEE International
IEEE International Symposium
International Symposium onon Smart
Symposium on Smart Electronic
Smart Electronic Systems
Electronic Systems (iSES)
Systems (iSES)
(iSES)

IPIP Core
Core Protection
Protection ofof Image
Image Processing
Processing Filters
Filters with
,3&RUH3URWHFWLRQRI,PDJH3URFHVVLQJ)LOWHUVZLWK with
Multi-Level
Multi-Level Encryption
Encryption and and Covert
Covert Steganographic
0XOWL/HYHO(QFU\SWLRQDQG&RYHUW6WHJDQRJUDSKLF
Steganographic
Security
Security Constraints
6HFXULW\&RQVWUDLQWV
Constraints
Aditya Anshul
$GLW\D$QVKXO Anirban Sengupta
$QLUEDQ6HQJXSWD
Aditya Anshul Anirban Sengupta
Computer Science and Engineering
&RPSXWHU6FLHQFHDQG(QJLQHHULQJ Computer Science and Engineering
&RPSXWHU6FLHQFHDQG(QJLQHHULQJ
Computer Science and Engineering Computer Science and Engineering
Indian Institute of Technology Indore, India
,QGLDQ,QVWLWXWHRI7HFKQRORJ\,QGRUH,QGLD Indian Institute of Technology Indore, India
,QGLDQ,QVWLWXWHRI7HFKQRORJ\,QGRUH,QGLD
2022 IEEE International Symposium on Smart Electronic Systems (iSES) | 979-8-3503-9922-6/22/$31.00 ©2022 IEEE | DOI: 10.1109/ISES54909.2022.00028

Indian Institute of Technology Indore, India Indian Institute of Technology Indore, India
[email protected]
SKG#LLWLDFLQ asengupt(@iiti.ac.in
DVHQJXSW#LLWLDFLQ
[email protected] asengupt(@iiti.ac.in

 
Abstract— Image processing-based hardware accelerators are
$EVWUDFW²,PDJHSURFHVVLQJEDVHGKDUGZDUHDFFHOHUDWRUVDUH counterfeiting, FORQLQJ
FRXQWHUIHLWLQJ cloning, LQVHUWLRQ
insertion RI of PDOLFLRXV
malicious ORJLFlogic, DQG
and
Abstract— Image processing-based hardware accelerators are counterfeiting, cloning, insertion of malicious logic, and
extensively used in consumer electronics devices and applications.
H[WHQVLYHO\XVHGLQFRQVXPHUHOHFWURQLFVGHYLFHVDQGDSSOLFDWLRQV fraudulent FODLP
IUDXGXOHQW claim RI of ,3IP RZQHUVKLS
ownership. 6HFXULQJ Securing WKHVH
these LPDJH
image
extensively used in consumer electronics devices and applications. fraudulent claim of IP ownership. Securing these image
Image SURFHVVLQJ
,PDJH processing LQWHOOHFWXDO
intellectual SURSHUW\
property (IP) cores VXFK
,3  FRUHV such DV
as EOXU
blur processing IP cores from the aforementioned hardware threats
SURFHVVLQJ,3FRUHVIURPWKHDIRUHPHQWLRQHGKDUGZDUHWKUHDWV
Image processing intellectual property (IP) cores such as blur processing IP cores from the aforementioned hardware threats
filter, sharpening filter, laplace edge detection filter, etc., are some
ILOWHUVKDUSHQLQJILOWHUODSODFHHGJHGHWHFWLRQILOWHUHWFDUHVRPH is therefore essential. Counterfeited IPs are unreliable as they
LVWKHUHIRUHHVVHQWLDO&RXQWHUIHLWHG,3VDUHXQUHOLDEOHDVWKH\
filter, sharpening filter, laplace edge detection filter, etc., are some is therefore essential. Counterfeited IPs are unreliable as they
essential IP cores used in these image processing-based hardware
HVVHQWLDO,3FRUHVXVHGLQWKHVHLPDJHSURFHVVLQJEDVHGKDUGZDUH are QRW
DUH not ULJRURXVO\
rigorously WHVWHGtested DQGand FDQ
can FDXVH
cause VHYHUH
severe GDPDJH
damage WR to WKH
the
essential IP cores used in these image processing-based hardware are not rigorously tested and can cause severe damage to the
accelerators. Security of these image processing IP cores is crucial
DFFHOHUDWRUV6HFXULW\RIWKHVHLPDJHSURFHVVLQJ,3FRUHVLVFUXFLDO end FRQVXPHU
HQG consumer, SURGXFWproduct GHVLJQHU
designer, FRQVXPHU
consumer HOHFWURQLFV
electronics (CE) &( 
accelerators. Security of these image processing IP cores is crucial end consumer, product designer, consumer electronics (CE)
as WKH\
DV they SHUIRUP
perform FULWLFDO
critical UHDOZRUOG
real-world IXQFWLRQV
functions VXFK
such DVas PHGLFDO
medical manufacturer DQG
PDQXIDFWXUHU and WKH the YHQGRU
vendor UHSXWDWLRQ
reputation. )RU For H[DPSOH
example,
as they perform critical real-world functions such as medical manufacturer and the vendor reputation. For example,
imagining, character recognition, etc. This paper presents a novel
LPDJLQLQJFKDUDFWHUUHFRJQLWLRQHWF7KLVSDSHUSUHVHQWVDQRYHO
imagining, character recognition, etc. This paper presents a novel counterfeited IPs causing excessive heat dissipation may lead
FRXQWHUIHLWHG,3VFDXVLQJH[FHVVLYHKHDWGLVVLSDWLRQPD\OHDG
hardware VHFXULW\
KDUGZDUH security WHFKQLTXH
technique IRU for ,3
IP FRUH
core SURWHFWLRQ
protection RIof LPDJH
image counterfeited IPs causing excessive heat dissipation may lead
hardware security technique for IP core protection of image to D
WR a VKRUWHU
shorter OLIH
life F\FOH
cycle RI of WKH
the GHYLFH
device, OHDNDJH
leakage RI of VHQVLWLYH
sensitive
processing ILOWHUVZLWK
SURFHVVLQJ filters with PXOWLOHYHO
multi-level HQFU\SWLRQ
encryption DQG
and VWHJDQRJUDSKLF
steganographic to a shorter life cycle of the device, leakage of sensitive
processing filters with multi-level encryption and steganographic information, incorrect computation of pixel values of an image
LQIRUPDWLRQLQFRUUHFWFRPSXWDWLRQRISL[HOYDOXHVRIDQLPDJH
security FRQVWUDLQWV
VHFXULW\ constraints. 7KLVThis SDSHU
paper SUHVHQWV
presents VLJQDWXUHEDVHG
signature-based information, incorrect computation of pixel values of an image
security constraints. This paper presents signature-based or YLGHR
RU video OHDGLQJ
leading WR to ZURQJ
wrong GLDJQRVLV
diagnosis LQ in FDVH
case RI of PHGLFDO
medical
hardware security methodology on image processing IP cores for
KDUGZDUHVHFXULW\PHWKRGRORJ\RQLPDJHSURFHVVLQJ,3FRUHVIRU or video leading to wrong diagnosis in case of medical
hardware security methodology on image processing IP cores for imaging, etc. Therefore, it is necessary to distinguish between
LPDJLQJHWF7KHUHIRUHLWLVQHFHVVDU\WRGLVWLQJXLVKEHWZHHQ
the ILUVW
WKH first WLPH
time. )XUWKHU
Further, WKH
the SDSHU
paper SUHVHQWV
presents HPEHGGLQJ
embedding KDUGZDUH
the first time. Further, the paper presents embedding hardware
hardware imaging, etc. Therefore, it is necessary to distinguish between
genuine LPDJH
JHQXLQH image SURFHVVLQJ
processing ILOWHUfilter ,3IP FRUHV
cores IURP
from FRXQWHUIHLWHG
counterfeited
security FRQVWUDLQWV
VHFXULW\ constraints JHQHUDWHG
generated ZLWK with PXOWLOHYHO
multi-level HQFU\SWLRQ
security constraints generated with multi-level encryption and
encryption DQGand genuine image processing filter IP cores from counterfeited
ones using detective control mechanism before integration into
RQHVXVLQJGHWHFWLYHFRQWUROPHFKDQLVPEHIRUHLQWHJUDWLRQLQWR
steganographic constraints on blur filter application using a high-
VWHJDQRJUDSKLFFRQVWUDLQWVRQEOXUILOWHUDSSOLFDWLRQXVLQJDKLJK
steganographic constraints on blur filter application using a high- ones using detective control mechanism before integration into
level synthesis framework. The article also reports the design cost
OHYHOV\QWKHVLVIUDPHZRUN7KHDUWLFOHDOVRUHSRUWVWKHGHVLJQFRVW hardware V\VWHPV
KDUGZDUH systems. 7KLV This SDSHU
paper SUHVHQWV
presents D a QRYHO
novel KDUGZDUH
hardware
level synthesis framework. The article also reports the design cost hardware systems. This paper presents a novel hardware
(in terms RI
LQ WHUPV of GHOD\
delay DQG
and DUHD
area)  RI
of WKH
the SURSRVHG
proposed PHWKRGRORJ\
methodology security technique for IP core protection of image processing
VHFXULW\WHFKQLTXHIRU,3FRUHSURWHFWLRQRILPDJHSURFHVVLQJ
(in terms of delay and area) of the proposed methodology security technique for IP core protection of image processing
corresponding WR
FRUUHVSRQGLQJ to GLIIHUHQW
different ILOWHUV
filters DQG
and Da ORZHU
lower SUREDELOLW\
probability RIof filters with multi-level encryption and steganographic security
ILOWHUVZLWKPXOWLOHYHOHQFU\SWLRQDQGVWHJDQRJUDSKLFVHFXULW\
corresponding to different filters and a lower probability of filters with multi-level encryption and steganographic security
coincidence, indicating the robustness of the proposed approach.
FRLQFLGHQFHLQGLFDWLQJWKHUREXVWQHVVRIWKHSURSRVHGDSSURDFK constraints. :LWK
FRQVWUDLQWV With WKH the KHOS
help RI of PXOWLOHYHO
multi-level HQFU\SWLRQ
encryption EDVHGbased
coincidence, indicating the robustness of the proposed approach. constraints. With the help of multi-level encryption based
generated VLJQDWXUH
JHQHUDWHG signature DXJPHQWHG
augmented ZLWK with VWHJDQRJUDSKLF
steganographic VHFXULW\
security
generated signature augmented with steganographic security
Keywords— ,PDJH
.H\ZRUGV² Image processing
SURFHVVLQJfilters, multi-level HQFU\SWLRQ
ILOWHUV PXOWLOHYHO encryption, ,3
IP constraints, LW
FRQVWUDLQWV it LV
is HDVLHU
easier WRto GLVWLQJXLVK
distinguish EHWZHHQbetween JHQXLQH
genuine DQG and
Keywords— Image processing filters, multi-level encryption, IP constraints, it is easier to distinguish between genuine and
cores, steganographic security constraints.
FRUHVVWHJDQRJUDSKLFVHFXULW\FRQVWUDLQWV counterfeited (fake)
FRXQWHUIHLWHG filter IPs which helps prove the ownership
IDNH ILOWHU,3VZKLFKKHOSVSURYHWKHRZQHUVKLS
cores, steganographic security constraints. counterfeited (fake) filter IPs which helps prove the ownership
conflict in case of fraudulent claim of IP ownership [2].
FRQIOLFWLQFDVHRIIUDXGXOHQWFODLPRI,3RZQHUVKLS>@
,
I. ,INTRODUCTION
1752'8&7,21 conflict in case of fraudulent claim of IP ownership [2].
I. INTRODUCTION Prior UHODWHG
3ULRU related ZRUNV works >@ [3] ± — >@[6] KDYH
have GLVFXVVHG
discussed DERXW about
Prior related works [3] — [6] have discussed about
Automation KDV
$XWRPDWLRQ has EHFRPH
become DQ an LQWHJUDO
integral SDUW
part RIof WKH
the FXUUHQW
current implementation of FPGA based image processing accelerators,
LPSOHPHQWDWLRQRI)3*$EDVHGLPDJHSURFHVVLQJDFFHOHUDWRUV
Automation has become an integral part of the current implementation of FPGA based image processing accelerators,
rapidly HYROYLQJ
UDSLGO\ evolving WHFKQRORJ\
technology. )URPFrom GHWHFWLRQ
detection DQG and UHFRJQLWLRQ
recognition [7] - [10] have proposed hardware accelerators for performing
>@>@KDYHSURSRVHGKDUGZDUHDFFHOHUDWRUVIRUSHUIRUPLQJ
rapidly evolving technology. From detection and recognition [7] - [10] have proposed hardware accelerators for performing
of OLFHQVH
RI license SODWHV
plates RIof YHKLFOHV
vehicles RQ on WROOV
tolls (character recognition), 
FKDUDFWHU UHFRJQLWLRQ convolutional neural network (CNN),
FRQYROXWLRQDOQHXUDOQHWZRUN while [11] has proposed
&11 ZKLOH>@KDVSURSRVHG
of license plates of vehicles on tolls (character recognition), convolutional neural network (CNN), while [11] has proposed
advanced
DGYDQFHG medical LPDJHU\
PHGLFDO imagery (image analysis),  ELRPHWULF
LPDJH DQDO\VLV biometric a QHXUDO
D neural QHWZRUNEDVHG
network-based PRGHO model WR to JHQHUDWH
generate LPSURYHG
improved YLVXDO
visual
advanced medical imagery (image analysis), biometric a neural network-based model to generate improved visual
fingerprinting, URERWLFV
ILQJHUSULQWLQJ robotics YLVLRQ
vision, DQG
and PLOLWDU\
military RSHUDWLRQV
operations WR to FDU
car images ZLWK
LPDJHV with WKH
the KHOS
help RIof OHDUQLQJ
learning VLJQDOsignal SURFHVVLQJ
processing SLSHOLQH
pipeline.
fingerprinting, robotics vision, and military operations to car images with the help of learning signal processing pipeline.
driving DXWRPDWLRQ
GULYLQJ automation, LPDJHimage SURFHVVLQJ
processing SOD\V plays D a FUXFLDO
crucial UROHrole Additionally, DSSURDFK
$GGLWLRQDOO\ approach >@ [12] KDVhas SHUIRUPHG
performed KDUGZDUH
hardware ,3 IP
driving automation, image processing plays a crucial role Additionally, approach [12] has performed hardware IP
everywhere. 7KH
HYHU\ZKHUH The PDLQ
main REMHFWLYH
objective RI of LPDJH
image SURFHVVLQJ
processing LV is WR
to watermarking RI
ZDWHUPDUNLQJ of VLJQDO
signal SURFHVVLQJ
processing NHUQHOV kernels XVLQJ
using ELQDU\
binary
everywhere. The main objective of image processing is to watermarking of signal processing kernels using binary
extract some useful information from an image. Various image
H[WUDFWVRPHXVHIXOLQIRUPDWLRQIURPDQLPDJH9DULRXVLPDJH encoding to protect against false claim of IP ownership threat.
HQFRGLQJWRSURWHFWDJDLQVWIDOVHFODLPRI,3RZQHUVKLSWKUHDW
extract some useful information from an image. Various image encoding to protect against false claim of IP ownership threat.
processing ILOWHUV
SURFHVVLQJ filters DUH
are UHTXLUHG
required WR to H[WUDFW
extract WKH the GHVLUHG
desired These DIRUHVDLG
7KHVH aforesaid DSSURDFKHV
approaches KDYH have QRWnot IRFXVHG
focused RQ on WKH
the VHFXULW\
security
processing filters are required to extract the desired These aforesaid approaches have not focused on the security
information IURP
LQIRUPDWLRQ from WKH
the LPDJH
image, VXFK
such DVas EOXU
blur ILOWHU
filter, VKDUSHQLQJ
sharpening aspects of image processing filter hardware IPs. Further none
DVSHFWVRILPDJHSURFHVVLQJILOWHUKDUGZDUH,3V)XUWKHUQRQH
information from the image, such as blur filter, sharpening aspects of image processing filter hardware IPs. Further none
filter, ODSODFH
ILOWHU laplace HGJH
edge GHWHFWLRQ
detection ILOWHU
filter, HWF
etc. 7KHVH
These ILOWHUV
filters SHUIRUP
perform of WKH
RI the SULRU
prior DSSURDFKHV
approaches H[SORLWHGexploited +/6 HLS IUDPHZRUN
framework IRU for
filter, laplace edge detection filter, etc. These filters perform of the prior approaches exploited HLS framework for
different IXQFWLRQV
GLIIHUHQW functions RQ on DQ
an LPDJH
image VXFK
such DVas EOXUULQJ
blurring, VKDUSHQLQJ
sharpening, generating secured filter hardware IP cores. In summary none
JHQHUDWLQJVHFXUHGILOWHUKDUGZDUH,3FRUHV,QVXPPDU\QRQH
different functions on an image such as blurring, sharpening, generating secured filter hardware IP cores. In summary none
embossment, edge detection, etc. These filters are designed as
HPERVVPHQWHGJHGHWHFWLRQHWF7KHVHILOWHUVDUHGHVLJQHGDV of WKH
RI the SUHYLRXV
previous SURSRVHG
proposed DSSURDFKHV
approaches KDYH have SUHVHQWHG
presented ,3 IP FRUH
core
embossment, edge detection, etc. These filters are designed as of the previous proposed approaches have presented IP core
dedicated LQWHOOHFWXDO
GHGLFDWHG intellectual SURSHUW\
property (IP) core. $OO
,3  FRUH All WKHVH
these RSHUDWLRQV
operations protection RI
SURWHFWLRQ of LPDJH
image SURFHVVLQJ
processing ILOWHUV filters XVLQJ
using PXOWLOHYHO
multi-level
dedicated intellectual property (IP) core. All these operations protection of image processing filters using multi-level
are computation intensive owing to the increasing VL]HRIWKH
DUHFRPSXWDWLRQLQWHQVLYHRZLQJWRWKHLQFUHDVLQJ size of the encryption and steganographic security constraints.
HQFU\SWLRQDQGVWHJDQRJUDSKLFVHFXULW\FRQVWUDLQWV
are computation intensive owing to the increasing size of the encryption and steganographic security constraints.
image and algorithm complexity. So, it is effective to harness
LPDJHDQGDOJRULWKPFRPSOH[LW\6RLWLVHIIHFWLYHWRKDUQHVV
image and algorithm complexity. So, it is effective to harness ,,
Il. 3PROPOSED WORK
52326(': 25.
the SRZHU
WKH power RI of KLJKOHYHO
high-level V\QWKHVLV
synthesis (HLS)+/6  IRUfor GHVLJQLQJ
designing WKHVHthese Il. PROPOSED WORK
the power of high-level synthesis (HLS) for designing these
complex image processing filters [1].
FRPSOH[LPDJHSURFHVVLQJILOWHUV>@ The WKUHDW
7KH threat PRGHO
model RI
of WKH
the SURSRVHG
proposed DSSURDFK
approach LV
is QXOOLI\LQJ
nullifying
complex image processing filters [1]. The threat model of the proposed approach is nullifying
Further, WKH
)XUWKHU the LQYROYHPHQW
involvement RI of PXOWLSOH
multiple RIIVKRUH
offshore HQWLWLHV
entities LQ in fraudulent FODLP
IUDXGXOHQW claim RI
of ,3
IP RZQHUVKLS
ownership XVLQJ
using SURSRVHG
proposed PXOWLOHYHO
multi-level
Further, the involvement of multiple offshore entities in fraudulent claim of IP ownership using proposed multi-level
the GHVLJQ
WKH design SURFHVV
process RI of WKHVH
these LPDJH
image SURFHVVLQJ
processing ,3 IP FRUHV
cores encryption and steganographic security constraints.
HQFU\SWLRQDQGVWHJDQRJUDSKLFVHFXULW\FRQVWUDLQWV
the design process of these image processing IP cores encryption and steganographic security constraints.
increases
LQFUHDVHV the
WKH risk RI
ULVN of YDULRXV
various hardware
KDUGZDUH threats
WKUHDWV like
OLNH
increases the risk of various hardware threats like

979-8-3503-9922-6/22/$31.00 ©2022
979-8-3503-9922-6/22/$31.00 ©2022 IEEE
IEEE 83
83
979-8-3503-9922-6/22/$31.00
DOI
DOI 10.1109/iSES54909.2022.00028
10.1109/iSES54909.2022.00028 ©2022 IEEE 83
DOI Authorized
10.1109/iSES54909.2022.00028
licensed use limited to: Amrita School of Engineering. Downloaded on October 04,2023 at 06:29:35 UTC from IEEE Xplore. Restrictions apply.
 Input: Scheduled dataflow graph of
,QSXW6FKHGXOHGGDWDIORZJUDSKRI Multi-Level Encryption
0XOWL/HYHO(QFU\SWLRQ
obfuscated 3*3LPDJHSURFHVVLQJ
REIXVFDWHG image processing
filters (blur, sharpening, and Laplace Key-bits
.H\ELWV Chosen mode of initial state matrix formation
&KRVHQPRGHRILQLWLDOVWDWHPDWUL[IRUPDWLRQ
ILOWHUV EOXUVKDUSHQLQJDQG/DSODFH Initial state matrix generation:
,QLWLDOVWDWHPDWUL[JHQHUDWLRQ
00
 Select 2 elements and skip subsequent 2.
6HOHFWHOHPHQWVDQGVNLSVXEVHTXHQW
edge detection filter) EDVHGRQJLYHQ
HGJHGHWHFWLRQILOWHU based on given ——PI Chose a subset S’ from set S of
&KRVHDVXEVHW6¶IURPVHW6RI OL
 Select 4 elements and skip subsequent 4.
6HOHFWHOHPHQWVDQGVNLSVXEVHTXHQW
resource constraints
UHVRXUFHFRQVWUDLQWV secret design data based on a
VHFUHWGHVLJQGDWDEDVHGRQD 10
 Select 8 elements and skip subsequent 8.
6HOHFWHOHPHQWVDQGVNLSVXEVHTXHQW
key value 1.
NH\YDOXH a
 Select 16 elements and skip subsequent 16.
6HOHFWHOHPHQWVDQGVNLSVXEVHTXHQW
y Vv
Generation of register allocation table Bit manipulation: Using forward AES-128 s-box, compute bit manipulation corresponding to
%LWPDQLSXODWLRQ8VLQJIRUZDUG$(6VER[FRPSXWHELWPDQLSXODWLRQFRUUHVSRQGLQJWR
*HQHUDWLRQRIUHJLVWHUDOORFDWLRQWDEOH
based on the scheduled data flow
EDVHGRQWKHVFKHGXOHGGDWDIORZ each element of state matrix.
HDFKHOHPHQWRIVWDWHPDWUL[
graph of corresponding image
JUDSKRIFRUUHVSRQGLQJLPDJH
processing filters
SURFHVVLQJILOWHUV Row diffusion: execute row
5RZGLIIXVLRQH[HFXWHURZ Key-bits
.H\ELWV Description of selected mode of row-diffusion
'HVFULSWLRQRIVHOHFWHGPRGHRIURZGLIIXVLRQ

— E—=E=E—EEEEESS
00
 Perform circular right shift operation by | element.
3HUIRUPFLUFXODUULJKWVKLIWRSHUDWLRQE\HOHPHQW
diffusion among the elements
GLIIXVLRQDPRQJWKHHOHPHQWV
01
 Perform circular right shift operation by 2 elements.
3HUIRUPFLUFXODUULJKWVKLIWRSHUDWLRQE\HOHPHQWV
of bit manipulated state matrix
RIELWPDQLSXODWHGVWDWHPDWUL[
N 10
 Perform circular right shift operation by 3 elements.
3HUIRUPFLUFXODUULJKWVKLIWRSHUDWLRQE\HOHPHQWV
based on key value 2.
EDVHGRQNH\YDOXH ll Perform circular right shift operation by 4 elements.
Extraction of secret design data from
([WUDFWLRQRIVHFUHWGHVLJQGDWDIURP  3HUIRUPFLUFXODUULJKWVKLIWRSHUDWLRQE\HOHPHQWV
register allocation table (set
UHJLVWHUDOORFDWLRQWDEOH of all
VHWRIDOO v
storage variables pairs allocated to
VWRUDJHYDULDEOHVSDLUVDOORFDWHGWR [Compute TRIFID cipher: Compute Trifid cipher on each unique alphabet of row-diffused state}
&RPSXWH75,),'FLSKHU&RPSXWH7ULILGFLSKHURQHDFKXQLTXHDOSKDEHWRIURZGLIIXVHGVWDWH
same register) 
VDPHUHJLVWHU matrix (with
PDWUL[ unique key value corresponding to different alphabet). 
ZLWKXQLTXHNH\YDOXHFRUUHVSRQGLQJWRGLIIHUHQWDOSKDEHW

Alphabetic substitution: Key-bits

.H\ELWV Description of rule to get digit equivalents
'HVFULSWLRQRIUXOHWRJHWGLJLWHTXLYDOHQWV

SSeS.
$OSKDEHWLFVXEVWLWXWLRQ

v Perform alphabetic substitution 

VT DweE F
3HUIRUPDOSKDEHWLFVXEVWLWXWLRQ
 DEF
to get corresponding equivalent
WRJHWFRUUHVSRQGLQJHTXLYDOHQW 010 = real
Hardware security constraints  _EFD_
+DUGZDUHVHFXULW\FRQVWUDLQWV digits based on key value 3 (aD
GLJLWVEDVHGRQNH\YDOXH O11 la-bre|
 _DEF_
generation based on mapping rule.
JHQHUDWLRQEDVHGRQPDSSLQJUXOH particular rule is applied on
SDUWLFXODUUXOHLVDSSOLHGRQ T00 (atbye

:
 DE F
output of trifid cipher). 
RXWSXWRIWULILGFLSKHU 101
 (a+b)
DE  *cF

Embedding of hardware security

(PEHGGLQJRIKDUGZDUHVHFXULW\
¥v
steganographic constraints into the Matrix transposition: Perform transpose operation on updated matrix.
0DWUL[WUDQVSRVLWLRQ3HUIRUPWUDQVSRVHRSHUDWLRQRQXSGDWHGPDWUL[

ee
VWHJDQRJUDSKLFFRQVWUDLQWVLQWRWKH
design of respective image processing
GHVLJQRIUHVSHFWLYHLPDJHSURFHVVLQJ
filter, generated from multi-level
ILOWHUJHQHUDWHGIURPPXOWLOHYHO Bvte concatenation: Key-bits
.H\ELWV Description of rule for byte concatenation
'HVFULSWLRQRIUXOHIRUE\WHFRQFDWHQDWLRQ
%\WHFRQFDWHQDWLRQ
encryption
HQFU\SWLRQ Concatenate each element of
&RQFDWHQDWHHDFKHOHPHQWRI 
VT &&&&
rn = + ~ 

update state matrix as per rule 010 tc&&&&
roy Ci. cy 

aec eee
XSGDWHVWDWHPDWUL[DVSHUUXOH
 &&&& 
decided using key value 4 and
GHFLGHGXVLQJNH\YDOXHDQG O11

(CO, C23. Cl)
&&&& 
convert them into
FRQYHUWWKHPLQWR 100
 (CO, C3. Ci. C2) 
&&&&
Output: Multi-level encryption and
2XWSXW0XOWLOHYHOHQFU\SWLRQDQG corresponding equivalent bytes.
FRUUHVSRQGLQJHTXLYDOHQWE\WHV 101 (C0, C3, C2, Cl) 
 &&&&
steganographic hardware security

|roo nasa
VWHJDQRJUDSKLFKDUGZDUHVHFXULW\
constraints secured obfuscated 3*3
FRQVWUDLQWVVHFXUHGREIXVFDWHG v
image processing filter
LPDJHSURFHVVLQJILOWHU Further, the byte sequence is truncated based on designer’s choice and used for corresponding
)XUWKHUWKHE\WHVHTXHQFHLVWUXQFDWHGEDVHGRQGHVLJQHU¶VFKRLFHDQGXVHGIRUFRUUHVSRQGLQJ
hardware security constraints generation.
KDUGZDUHVHFXULW\FRQVWUDLQWVJHQHUDWLRQ

Fig.1. Flow-chart of proposed methodology for securing image processing filters using multi-level encryption and steganographic hardware
)LJ)ORZFKDUWRISURSRVHGPHWKRGRORJ\IRUVHFXULQJLPDJHSURFHVVLQJILOWHUVXVLQJPXOWLOHYHOHQFU\SWLRQDQGVWHJDQRJUDSKLFKDUGZDUH
security constraints.
VHFXULW\FRQVWUDLQWV
The SURSRVHG
7KH proposed PHWKRGRORJ\
methodology IRU for VHFXULQJ
securing LPDJH image SURFHVVLQJ
processing value. Subsequently each element of the set is converted into
YDOXH6XEVHTXHQWO\HDFKHOHPHQWRIWKHVHWLVFRQYHUWHGLQWR
filters
ILOWHUV IP
,3 cores
FRUHV using
XVLQJ multi-level
PXOWLOHYHO encryption
HQFU\SWLRQ and
DQG its KH[DGHFLPDO
LWV hexadecimal YDOXHV
values IRUfor LQLWLDO
initial VWDWH
state PDWUL[
matrix FUHDWLRQ
creation EDVHG
based
steganographic VHFXULW\
VWHJDQRJUDSKLF security FRQVWUDLQWV
constraints KDV has EHHQ
been LOOXVWUDWHG
illustrated LQ in on the IP vendor selected key value. Further, bit-manipulation
RQWKH,3YHQGRUVHOHFWHGNH\YDOXH)XUWKHUELWPDQLSXODWLRQ
Fig.1. First, the kernel function matrix of the respective image
)LJ)LUVWWKHNHUQHOIXQFWLRQPDWUL[RIWKHUHVSHFWLYHLPDJH is SHUIRUPHG
LV performed XVLQJ
using D a IRUZDUG
forward $(6
AES-128 VER[ s-box (substitution
VXEVWLWXWLRQ
processing filter is taken. Then, based on the kernel function, a
SURFHVVLQJILOWHULVWDNHQ7KHQEDVHGRQWKHNHUQHOIXQFWLRQD box). 6XEVHTXHQWO\URZGLIIXVLRQLVSHUIRUPHGEDVHGRQDQ,3
ER[ Subsequently, row diffusion is performed based on an IP
generic
JHQHULF function
IXQFWLRQ corresponding
FRUUHVSRQGLQJ to
WR the
WKH output
RXWSXW pixel
SL[HO vendor FKRVHQ
YHQGRU chosen NH\
key YDOXH
value IRUfor URZ
row GLIIXVLRQ
diffusion FRUUHVSRQGLQJ
corresponding WR to
computation RI
FRPSXWDWLRQ of WKH
the UHVSHFWLYH
respective LPDJH image SURFHVVLQJ
processing ILOWHU filter each URZ
HDFK row LQ
in D a ELWPDQLSXODWHG
bit-manipulated VWDWH state PDWUL[
matrix. $IWHU
After WKDW
that,
application is generated. The generic transfer function is then
DSSOLFDWLRQLVJHQHUDWHG7KHJHQHULFWUDQVIHUIXQFWLRQLVWKHQ alphabetic VXEVWLWXWLRQ
DOSKDEHWLF substitution LV is SHUIRUPHG
performed FRUUHVSRQGLQJ
corresponding WR to DOO
all
converted LQWR
FRQYHUWHG into FRQWURO
control GDWDdata IORZ
flow JUDSK
graph (CDFG) &')*  RI of WKDW
that alphabets present in the state matrix with the help of TRIFID
DOSKDEHWVSUHVHQWLQWKHVWDWHPDWUL[ZLWKWKHKHOSRI75,),'
particular filter application. The obtained CDFG is scheduled
SDUWLFXODUILOWHUDSSOLFDWLRQ7KHREWDLQHG&')*LVVFKHGXOHG cipher FRPSXWDWLRQ
FLSKHU computation DQG and ,3IP YHQGRU
vendor VHOHFWHG
selected NH\ key YDOXHV
values IRU for
based RQ
EDVHG on WKHthe UHVRXUFH
resource FRQVWUDLQWV
constraints VHOHFWHG
selected E\ by WKH
the GHVLJQHU
designer. alphabetic VXEVWLWXWLRQ
DOSKDEHWLF substitution. )RU For FRPSXWLQJ
computing WKH the 75,),'
TRIFID FLSKHU
cipher,
After WKDW
$IWHU that, WKH the UHJLVWHU
register DOORFDWLRQ
allocation WDEOH table LV is JHQHUDWHG
generated again, a TRIFID cipher key value is selected by the IP vendor.
DJDLQD75,),'FLSKHUNH\YDOXHLVVHOHFWHGE\WKH,3YHQGRU
corresponding to the target filter application with the help of a
FRUUHVSRQGLQJWRWKHWDUJHWILOWHUDSSOLFDWLRQZLWKWKHKHOSRID Matrix transposition is performed after alphabetic substitution
0DWUL[WUDQVSRVLWLRQLVSHUIRUPHGDIWHUDOSKDEHWLFVXEVWLWXWLRQ
scheduled GDWD
VFKHGXOHG data IORZ
flow JUDSK
graph (SDFG).
6')*  1RZ Now, VWHJDQRJUDSKLF
steganographic to LQWURGXFH
WR introduce PRUH
more GLYHUVLILFDWLRQ
diversification LQ in WKH
the ILQDO
final VLJQDWXUH
signature
security constraints (secret
VHFXULW\FRQVWUDLQWV information) DUHH[WUDFWHGIURPWKH
VHFUHWLQIRUPDWLRQ are extracted from the generation. )LQDOO\
JHQHUDWLRQ Finally, DOO
all E\WHV
bytes FRUUHVSRQGLQJ
corresponding WR to HDFK
each FROXPQ
column
register DOORFDWLRQ
UHJLVWHU allocation WDEOH table RI
of WKH
the WDUJHW
target ILOWHU
filter DSSOLFDWLRQ
application. +HUH
Here, are FRQFDWHQDWHG
DUH concatenated LQWRinto Da VLQJOH
single VWULQJ
string EDVHG
based RQ on WKH
the ,3
IP YHQGRU
vendor
the SDLU
WKH pair RI
of VHFXULW\
security FRQVWUDLQWV
constraints (Zi, Zk)N  ZKLFK
=L = which FDQ can EH
be DGGHG
added selected key YDOXH
VHOHFWHGNH\ value IRUE\WHFRQFDWHQDWLRQDQGWKHQFRQYHUWHG
for byte concatenation and then converted
between the storage variable of the same color is generated.
EHWZHHQWKHVWRUDJHYDULDEOHRIWKHVDPHFRORULVJHQHUDWHG into binary equivalents. The final obtained string is the multi-
LQWRELQDU\HTXLYDOHQWV7KHILQDOREWDLQHGVWULQJLVWKHPXOWL
The secret design data is represented by a VHW
7KHVHFUHWGHVLJQGDWDLVUHSUHVHQWHGE\D set µ6¶FRPSULVLQJ
‘S’ comprising level encryption and steganography security constraints-based
OHYHOHQFU\SWLRQDQGVWHJDQRJUDSK\VHFXULW\FRQVWUDLQWVEDVHG
of elements indicating ‘indices value (i,LN
RIHOHPHQWVLQGLFDWLQJµLQGLFHVYDOXH k) RIVWRUDJHYDULDEOH
of storage variable signature, ZKLFK
VLJQDWXUH which LV is IXUWKHU
further FRQYHUWHG
converted LQWRinto KDUGZDUH
hardware VHFXULW\
security
pair (Zi,
SDLU =L=Z,)N DVVLJQHGWRVDPHUHJLVWHUFRORULQWKH5$77KXV
assigned to same register/color in the RAT. Thus, constraints EDVHG
FRQVWUDLQWV based RQ on PDSSLQJ
mapping UXOHVrules RI of ,3
IP YHQGRU
vendor DQG and
each HOHPHQW
HDFK element RI of WKH
the VHW
set LV
is D
a SDLU
pair RI
of WZR
two GLJLWV
digits RI
of WKH
the LQGLFHV
indices embedded into the target filter IP core design. The multi-level
HPEHGGHGLQWRWKHWDUJHWILOWHU,3FRUHGHVLJQ7KHPXOWLOHYHO

84
84

Authorized licensed use limited to: Amrita School of Engineering. Downloaded on October 04,2023 at 06:29:35 UTC from IEEE Xplore. Restrictions apply.
 encryption RYHU
HQFU\SWLRQ over VHFUHW
secret VWHJDQRJUDSKLF
steganographic LQIRUPDWLRQ
information LQFUHDVHV
increases 111
the UREXVWQHVV
WKH robustness DQG
and VWUHQJWK
strength RI
of WKH
the JHQHUDWHG
generated VLJQDWXUH
signature. 1
Moreover, a significant variation of key value in the proposed
0RUHRYHUDVLJQLILFDQWYDULDWLRQRINH\YDOXHLQWKHSURSRVHG Kernel = *! 1 1 1 |, 6LPLODUO\
Similarly, WKH
the NHUQHO
kernel IXQFWLRQ
function RI
of
Blur 9
3Poo
 3Por
 3Por
 3Pio
 3Pir
 3Pis
 3Poo
 3Poi
 3P22
   3Poo
1/9 3Por  3Pos
 3Pi
 3Pi2
 Pis
3 3P21
 3Poo
 3py;
 1/9 
 111
= = =
ZO
= Zl =
= Z2 73 =
= Z4 = Z5Z6= =2728 = =Z9 = zll =
zI90 = Z12 =
Z13Z14Z15
Z16
=
Z17
Z18
=
Z19
= 3*3VKDUSHQLQJDQG
sharpening, and 3*3/DSODFHILOWHULVVKRZQEHORZ
Laplace filter is shown below.
5R < 9V /, $ Gr
*U 6 Py / 0 -l O -1 -1 -1
:
* B % 3
P 2 7T . & %
, %O 0
& Kernel = -1 4 -1 Kern = -1 9 -!1 
laplace e el ss arpening
$

 0 -!l O -1 -1 -1
= & Based on the blur filter'sVNHUQHOIXQFWLRQWKHWUDQVIHUIXQFWLRQ
%DVHGRQWKHEOXUILOWHU kernel function, the transfer function


$ corresponding to the first two-pixel computation output value
FRUUHVSRQGLQJWRWKHILUVWWZRSL[HOFRPSXWDWLRQRXWSXWYDOXH
= & is formulated as O;DQG2
LVIRUPXODWHGDV2 and Op.
 
$

&
2Oo =  >[(Poo*(1/9))
3   + (Por*(1/9))
3    + (Po2*(1/9))]
3  @ > +[(P10*(1/9))
3    +
(Pi*(1/9)) + (Pr*(1/9))] +[(P20*(1/9)) + (Pai*C1/9)) +
=
3    3  @ > 3    3   

 $ 3  @ qd)
(P22*(1/9))]  
= & 2O1 =  >[(Por*(1/9))
3   + (Po2*(1/9))
3    + (Pos*(1/9))]
3  @ > +[(Pui*C1/9))
3    +
 $
(Pi2*(1/9))
3    + (Pis*(1/9))]
3  @ > +[(Pai*(1/9))
3    + (P22*(1/9))
3    +
=  3  @ (2)
(P23*(1/9))] 
&

=  $
$IWHU formulating WKH
After IRUPXODWLQJ the WUDQVIHU
transfer IXQFWLRQ
function, WKH the VFKHGXOHG
scheduled GDWD data

& flow JUDSK
IORZ graph (SDFG)
6')*  RIof WKH
the EOXU
blur ILOWHU
filter DSSOLFDWLRQ
application LV is JHQHUDWHG
generated

$ with WKH
ZLWK the KHOS
help RI
of GHVLJQHUVSHFLILF
designer-specific UHVRXUFHresource FRQVWUDLQWV
constraints (for IRU

= & demonstration, 
GHPRQVWUDWLRQ 1 DGGHU
adder,  1 VXEWUDFWRU
subtractor, DQG and  2 PXOWLSOLHUV
multipliers DUH are
 $
selected).7KH6')*FRUUHVSRQGLQJWRWKHEOXUILOWHUEDVHGRQ
VHOHFWHG The SDFG corresponding to the blur filter based on
 = equations (1)
HTXDWLRQV and (2)
 DQG have been shown in Fig. 2.
 KDYHEHHQVKRZQLQ)LJ
&

0
 $ B. ,3FRUHSURWHFWLRQRILPDJHSURFHVVLQJILOWHUVXVLQJ
% IP core protection of image processing filters using

=
= & proposed approach
SURSRVHGDSSURDFK
$  Register allocation table (RAT)
5HJLVWHUDOORFDWLRQWDEOH corresponding to the target
5$7 FRUUHVSRQGLQJWRWKHWDUJHW

2 & image SURFHVVLQJ
LPDJH processing ILOWHU filter LV is JHQHUDWHG
generated IURP from WKH the UHVSHFWLYH
respective
= SDFG. RAT corresponding to SDFG of blur filter is given in
6')*5$7FRUUHVSRQGLQJWR6')*RIEOXUILOWHULVJLYHQLQ
 $
 Table 
7DEOH 1 (black-
EODFN DQGand LQGLJRFRORUHG
indigo-colored UHJLVWHUV registers DUH are LQLWLDO
initial
= &

positions RI
SRVLWLRQV of UHJLVWHUV
registers DQG and UHGred FRORUHG
colored UHJLVWHUV
registers DUH are ILQDO
final
$
 position DIWHU
SRVLWLRQ after VLJQDWXUH
signature HPEHGGLQJembedding).  7KHQ Then, FRQILGHQWLDO
confidential
= &
(secret) data LV
VHFUHW  GDWD is H[WUDFWHG
extracted IURP from WKH the UHVSHFWLYH
respective 5$7 RAT. $OO All WKH
the
pairs of storage variables (Zi, =L=Zx)N DOORFDWHGWRWKHVDPHFRORU
allocated to the same color'sV

$
 SDLUVRIVWRUDJHYDULDEOHV
=
& registers DUH
UHJLVWHUV are OLVWHG
listed. )XUWKHU
Further, LQGLFHV
indices YDOXHV
values (i,L Nk)  RI
of VWRUDJH
storage
=
$ variable SDLU
YDULDEOH pair (Zi,=L =Zx)N  DUH
are FRQYHUWHG
converted WR to WKHLU
their KH[DGHFLPDO
hexadecimal
 
& equivalents. 7KH
HTXLYDOHQWV The OLVW
list RI
of ILQDO
final JHQHUDWHG
generated FRQILGHQWLDO
confidential GDWD data LQin
 $ hexadecimal IRUP
KH[DGHFLPDO form LV is DOVR
also NQRZQ
known DV as VWHJDQRJUDSKLF
steganographic VHFXULW\ security

= &
constraints (secret
FRQVWUDLQWV VHFUHW GDWD data). 1RZ
Now, PXOWLOHYHO
multi-level HQFU\SWLRQ
encryption LV is

performed RQ
SHUIRUPHG on FRQILGHQWLDO
confidential GDWD data WRto JHQHUDWH
generate WKH the VLJQDWXUH
signature DQG and
 $ corresponding hardware security constraints. The steps of the
FRUUHVSRQGLQJKDUGZDUHVHFXULW\FRQVWUDLQWV7KHVWHSVRIWKH
proposed multi-level encryption are explained below.
= &
SURSRVHGPXOWLOHYHOHQFU\SWLRQDUHH[SODLQHGEHORZ
 0 (a). Initial VWDWH
D  ,QLWLDO state PDWUL[
matrix JHQHUDWLRQ
generation: $ A VWDWH
state PDWUL[
matrix LV is IRUPHG
formed
= & based RQ
EDVHG on WKH
the ,3
IP YHQGRU
vendor FKRVHQ chosen NH\ key YDOXHV
values IRU for VWDWH
state PDWUL[
matrix
formation (key
IRUPDWLRQ value 
NH\ YDOXH 1 DV
as PHQWLRQHG
mentioned LQ in )LJ
Fig. 1). 7KH
The VWDWH
state
2O1 matrix should consist of a maximum of 4 elements per row. A
PDWUL[VKRXOGFRQVLVWRIDPD[LPXPRIHOHPHQWVSHUURZ$
Fig.2. Scheduled data flow graph of 3*3EOXUILOWHUZLWK
)LJ6FKHGXOHGGDWDIORZJUDSKRI blur filter with 1(+),
 1(-) and 2(*) 
 DQG
two-bit NH\
WZRELW key KDV
has EHHQ
been XVHGused KHUH
here IRU for FKRRVLQJ
choosing WKH the PRGH
mode RI of
as resource constraints
DVUHVRXUFHFRQVWUDLQWV
initial state matrix generation.
LQLWLDOVWDWHPDWUL[JHQHUDWLRQ
multi-level HQFU\SWLRQ
PXOWLOHYHO encryption LQFUHDVHV
increases WKH
the VHFXULW\
security DQG
and PDNHV
makes WKH
the
(b). Bit-manipulation: Each element of the initial state matrix
E %LWPDQLSXODWLRQ(DFKHOHPHQWRIWKHLQLWLDOVWDWHPDWUL[
hardware VHFXULW\
KDUGZDUH security PHWKRGRORJ\
methodology LPPXQH
immune DQGand FRPSOH[
complex IRU
for
is VXEVWLWXWHG
LV substituted ZLWK
with LWVits HTXLYDOHQW
equivalent VXEVWLWXWHV
substitutes XVLQJusing D a IRUZDUG
forward
regeneration from an attacker’s perspective.
UHJHQHUDWLRQIURPDQDWWDFNHU¶VSHUVSHFWLYH
AES-128 VER[
$(6 s-box. ,W It LV
is WKH
the QRQOLQHDU
non-linear PDQLSXODWLRQ
manipulation RI of HDFK
each
A. ,PDJHSURFHVVLQJILOWHUIXQFWLRQV
$ Image processing filter functions matrix element. It imparts confusion.
PDWUL[HOHPHQW,WLPSDUWVFRQIXVLRQ
The mathematical kernel function of a 3*3EOXUILOWHULV
7KHPDWKHPDWLFDONHUQHOIXQFWLRQRID blur filter is: (c) Row GLIIXVLRQ
F  5RZ diffusion: 5RZGLIIXVLRQ
Row-diffusion LV is LPSOHPHQWHG
implemented SRVWELW post-bit-
 manipulation based on the IP vendor chosen key value for row
PDQLSXODWLRQEDVHGRQWKH,3YHQGRUFKRVHQNH\YDOXHIRUURZ
diffusion (key
GLIIXVLRQ value 2). 6LQFH
NH\ YDOXH Since WKHUH
there DUHare IRXU
four PRGHV
modes RI of URZ
row-
diffusion, so, the key value is represented with 2-bits. Further
GLIIXVLRQVRWKHNH\YDOXHLVUHSUHVHQWHGZLWKELWV)XUWKHU

85
85
Authorized licensed use limited to: Amrita School of Engineering. Downloaded on October 04,2023 at 06:29:35 UTC from IEEE Xplore. Restrictions apply.
        7$%/(
TABLE, I
    5(*,67(5$//2&$7,217$%/(2)
REGISTER ALLOCATION TABLE OF 3*3
%/85),/7(5'(3,&7(',1
BLUR FILTER DEPICTED IN),* 2
FIG.
&6 5HG 5  *UHHQ ,QGLJR %OXH <HOORZ %ODFN 9LROHW 3LQN /LPH 2OLYH $TXD
Aqua 7HDO *UD\ 0DURRQ 6LOYHU .KDNL /DYHQGHU &ULPVRQ :KHDW %HLJH
* , %/  < % 9 3 /,  2 $ 7 * 0 6 . / & :  % 
 =ZO =Zi = = = = = = = = =
Z10 = = = = = = = = =
 = = = = = = = = = = = = = = = = = = = =
 ==
Z20/Z21 ==
2Z21/Z20   = = = = = = = = = = = = = = = =
 =
222 =
Z22   = = = = = = = = = = = = = = = =
 == ==     = = = = = = = = = = = = = =
 ==
Z22/Z23 ==
Z23/Z22 = =     = = = = = = = = = = = =
 ==
Z22/Z25 ==
Z25/Z22       = = = = = = = = = = = =
 = =       = = = = = = = = = = = =
 =
227         = = = = = = = = = = =
 ==
Z28/Z29 ==
Z29/Z28           = = = = = = = =
 ==
Z30/Z29 ==
2Z29/Z30             = = = = = =
 =              = = = = = =
 = =               = = = =
 = =
Z32 =                = =
 = =                 = =
 =                  = =
 =                   =
 =                   


WKHILQDO
the final NH\ key YDOXH
value IRUURZGLIIXVLRQFRUUHVSRQGLQJWRDVWDWH
for row-diffusion corresponding to a state UHVRXUFHFRQVWUDLQWV
resource constraints IRUGHPRQVWUDWLRQZHKDYHDVVXPHGRQH
(for demonstration, we have assumed one
matrix LVis  2*L
PDWUL[ (where /L LVis WKH
/ ZKHUH the QXPEHU
number RI of URZV
rows LQ in WKH
the VWDWH
state adder (+),RQHVXEWUDFWRU
DGGHU one subtractor (-), and two multipliers (*)).:HKDYH
DQGWZRPXOWLSOLHUV We have
PDWUL[
matrix). considered WKH
FRQVLGHUHG the GDWD
data IORZ
flow JUDSKgraph RI of EOXU
blur ILOWHU
filter ZLWK
with ORRS
loop
G(d)&RPSXWLQJ75,),'FLSKHU75,),'FLSKHUKHOSVWRLPSDUW
Computing TRIFID cipher: TRIFID cipher helps to impart XQUROOLQJ
unrolling FRQFXUUHQW
(concurrent FRPSXWDWLRQ
computation IRU for WZRSL[HO
two-pixel YDOXHVvalues) DQG
and
FRQIXVLRQ
confusion DQG and GLIIXVLRQ
diffusion E\ by LPSOHPHQWLQJ
implementing WKUHH three SURSHUWLHV
properties: WUHHKHLJKWWUDQVIRUPDWLRQ
tree-height-transformation. )XUWKHU Further, WKH the UHJLVWHU
register DOORFDWLRQ
allocation
IUDFWLRQDWLRQ
fractionation, VXEVWLWXWLRQ
substitution, DQG
and __ WUDQVSRVLWLRQ
transposition. 7KH
The WDEOH
table LVis JHQHUDWHG
generated EDVHG based RQ on WKH
the 6')*
SDFG RI of WKH
the EOXU
blur ILOWHU
filter
LPSOHPHQWDWLRQRIWKH75,),'FLSKHULVH[SODLQHGLQGHWDLOLQ
implementation of the TRIFID cipher is explained in detail in DSSOLFDWLRQ&RYHUW
application. Covert VHFUHW (secret)GDWDLVH[WUDFWHGIURPWKHUHJLVWHU
data is extracted from the register
the QH[W
WKH next VXEVHFWLRQ
subsection. 7KH The WRWDO
total NH\
key VL]H
size XVHG
used LQin WKHthe 75,),'
TRIFID allocation WDEOH
DOORFDWLRQ table DVas H[SODLQHG
explained LQin WKH the SUHYLRXV
previous VXEVHFWLRQ
subsection. $OOAll
FLSKHU
cipher WKHUH(there DUH
are  27 XQLTXH
unique FKDUDFWHUV
characters LQ in WKH
the DOSKDEHW
alphabet, storage YDULDEOH
VWRUDJH variable SDLUVpairs =(Zi,L =Zx) allocated RQ
N  DOORFDWHG on WKHthe VDPH
same FRORU
color
including "$’’)LV
LQFOXGLQJ´ is: register are generated. The list of such pairs is:
UHJLVWHUDUHJHQHUDWHG7KHOLVWRIVXFKSDLUVLV
ሺ͓‘ˆ—‹“—‡ƒŽ’Šƒ„‡–•ሻ
(# of unique alphabets) ‫« כ‬᪬ ݈‫݃݋‬ ଶ ሺʹ͹Ǩሻ ᪭
-log2(27!) 4 (3) 6S=^ {(0,20),
   (0,22), 
(0,26), 
(0,27), 
(0,28), 
(0,30), 
(0,31), 
(0,35),
 (0,36), 
 (0,37), 
(20,22),  (20,26), 
(20,27), 
(20,28), (20,30),
H(e).$OSKDEHWLFVXEVWLWXWLRQ$OOXVHGXQLTXHDOSKDEHWVLQVWDWH
Alphabetic substitution: All used unique alphabets in state (20,31), 
 (20,35), 
(20,36), 
(20,37), 
(22,26), 
(22,27), 
(22,28),
matrix LVis FRQYHUWHG
PDWUL[ converted WR to WKHLU
their HTXLYDOHQW
equivalent GLJLW
digit, EDVHG
based RQ on WKH
the (22,30), 
 (22,31), 
(22,35), 
(22,36), 
(22,37), 
(26,27), 
(26,28),
RXWSXW
output VWDWH
(state) RI
of 75,),'
TRIFID FLSKHUcipher DQGand ,3
IP YHQGRU
vendor VHOHFWHG
selected NH\ key (26,30), 
 (26,31), 
(26,35), 
(26,36), 
(26,37), 
(27,28), 
(27,30),
value IRU
YDOXH for DOSKDEHWLF
alphabetic VXEVWLWXWLRQ
substitution NH\ (key YDOXH
value  3).  $A SDUWLFXODU
particular (27,31), 
 (27,35), 
(27,36), 
(27,37), 
(28,30), 
(28,31), 
(28,35),
digit FRQYHUVLRQ
GLJLW conversion UXOH rule LVis GHWHUPLQHG
determined EDVHGbased RQ on NH\
key YDOXH
value 3 IRU
for (28,36), 
 (28,37), 
(30,31), 
(30,35), 
(30,36), 
(30,37), 
(31,35),
DOSKDEHWLFVXEVWLWXWLRQZKLFKZLOOPDSDVSHFLILFDOSKDEHWWRD
alphabetic substitution, which will map a specific alphabet to a (31,36), 
 (31,37), 
(34,36), 
(35,37), 
(36,37), (1,21), 
(1,23),
SDUWLFXODUGLJLW
particular digit. (1,25), 
 (1,29), 
(1,32), 
(1,34), 
(21,23), 
(21,25), 
(21,29), 
(21,32),
I(). 0DWUL[
Matrix 7UDQVSRVLWLRQ
Transposition: 3RVW Post DOSKDEHWLF
alphabetic VXEVWLWXWLRQ
substitution, WKH the (21,34), 
 (23,25), 
(23,29),  (23,34), 
(23,32),  (25,29), 
(25,32),
final updated matrix is transposed.
ILQDOXSGDWHGPDWUL[LVWUDQVSRVHG (25,34), 
 (29,32), 
(29,34), 
(32,34), 
(2,24), 
(2,33), 
(33,24)}.`
J(g). %\WH concatenation: $OO
Byte FRQFDWHQDWLRQ All WKHthe HOHPHQWV
elements RI of WKHthe ILQDO
final TABLE ,,II
7$%/( TABLE ,,,
7$%/( III TABLE,9
7$%/( IV
*GENERATED INITIAL 67$7(0$75,;$)7(5%,7
(1(5$7(',1,7,$/ STATE MATRIX AFTERBIT 67$7(0$75,;$)7(5
STATE MATRIX AFTER
generated PDWUL[
JHQHUDWHG matrix DUH are FRQFDWHQDWHG
concatenated EDVHG based RQ on WKH the UXOHV
rules
STATE MATRIX
67$7(0$75,; 0$1,38/$7,21
MANIPULATION 6(S-BOX%2;  ROW_DIFFUSION
52:',))86,21
corresponding to IP vendor chosen key for byte concatenation
FRUUHVSRQGLQJWR,3YHQGRUFKRVHQNH\IRUE\WHFRQFDWHQDWLRQ 0S | 
 07 | %OB | &
OC} % J6B | &
CS | % 2B | )(
FE|]||C5
& | % 2B | )(
FE | %
6B
(key value 4).
NH\YDOXH Further, the values are converted to their binary
)XUWKHUWKHYDOXHVDUHFRQYHUWHGWRWKHLUELQDU\ 57_| %
 5B | &SC | '
SD|} % 5B | 39 | $ 4A | &
4C]] &/4C | %5B | 
39 | $
4A
HTXLYDOHQWV7KHWRWDONH\VL]HIRUE\WHFRQFDWHQDWLRQLV
equivalents. The total key size for byte concatenation is: 7C_| '
& 7D | 71 | 
76 10 | ))
FF | $ A3 | 
38}, {10 | ))
FF | $
A3 | 
38
(total number of ‘Ž—•‹–”ƒ•’‘•‡†ƒ–”‹šሻ
ሺ–‘–ƒŽ—„‡”‘ˆ columns in trasposed matrix) ‫* כ‬ CD | &
&' Cl | 16 | '
DI} %' JBD | 
78 | % B4 | (
3E}]|B4
% | ( 3E | %'
BD | 
78
݈‫݃݋‬logଶ ሺ–‘–ƒŽ’‘••‹„Ž‡™ƒ›•‘ˆ
(total possible ways of ‘ ƒ–ƒ–‹‘ሻ (4) 1A | ( 1E | 12 | 14} $ JA2 | 
72 | & C9 | )$
FAJT )$IFA | $
A2 | 72 | &
C9
$
concatnation)
64 | $
 8A | (8E | 8241  ]43 | (
7E | 19 | 134] 
]43 | (7E | 19 | 13

E2_ | (
( E4 | 24 |} 
23 98 | 
 69 |  36 | 
26}||69
 |  36 | 
26 | 
98
)LQDOO\WKHVWHJDQRJUDSKLFVHFXULW\FRQVWUDLQWVDQGPXOWLOHYHO
Finally, the steganographic security constraints and multi-level
encryption-based VLJQDWXUH
HQFU\SWLRQEDVHG signature DUH are WUXQFDWHG
truncated EDVHG based RQ on WKH
the 1RZ
Now, DOO all RIof WKHP
them DUH are FRQYHUWHG
converted LQWR into WKHLU
their KH[DGHFLPDO
hexadecimal
designer's
GHVLJQHU choice DQG
V FKRLFH and FRQYHUWHG
converted LQWR into LWV its FRUUHVSRQGLQJ
corresponding HTXLYDOHQWV
equivalents IRU for SHUIRUPLQJ
performing PXOWLOHYHO
multi-level HQFU\SWLRQ
encryption. 7KH The ILQDO
final
KDUGZDUHVHFXULW\FRQVWUDLQWVEDVHGRQWKHPDSSLQJUXOHRI,3
hardware security constraints based on the mapping rule of IP
OLVWRIJHQHUDWHGVHFUHWGDWDSDLUVLQWKHLUKH[DGHFLPDOIRUPLV
list of generated secret data pairs in their hexadecimal form is:
vendor LIHQFRGLQJELWRIVLJQDWXUHLVµ¶WKHQHPEHGDQHGJH
YHQGRU (if encoding bit of signature is ‘0’ then embed an edge 6S = ^ {(0,5),
   (0,7), %(0,B), &
(0,C), '
(0,D), (0,1), 
(0,5), 
(0,6), 
(5,7),
EHWZHHQ
between HYHQ (even, HYHQ
even) VWRUDJH
storage YDULDEOH
variable SDLU
pair, RWKHUZLVH
otherwise HPEHG embed (5,B), (5,C), (5,D), (5,1), (5,6), (5,7), (7,B), (7,0), (7,D), (7,1),
%  &  '        %  &  '  
an edge between RGGRGG
DQHGJHEHWZHHQ (odd, odd)VWRUDJHYDULDEOHSDLU
storage variable pair)DQGIXUWKHUand further (7,6), %&
 (B,C), %'(B,D), % (B,1), %
(B,6), &'
(C,D), & (C.D), &(C,6), '
(D,1),
embedded into the design flow of the target IP core.
HPEHGGHGLQWRWKHGHVLJQIORZRIWKHWDUJHW,3FRUH (D,6), 
' (1,6), 
(1,8), $(1,A), (
(1,E), 
(1,2), 
(1,4), 
(6,8), $
(6,4), (
(6,E),
(6,2), (6,4), (8,A), (8,E), (8,2), (8,4), (A,E), (A,2), (A,4), (E,2),
    $  (      $(  $  $  (
C. 'HPRQVWUDWLRQRISURSRVHGDSSURDFKRQEOXUILOWHU,3FRUH
& Demonstration of proposed approach on blur filter IP core
(E,4), 
( (2,4), 
(2,3), 3,9)}.
`
$VH[SODLQHGLQWKHSUHYLRXVVHFWLRQ)LJUHSUHVHQWVWKH
As explained in the previous section, Fig. 2 represents the $IWHUWKHJHQHUDWLRQRIVHFUHWGDWD
After the generation of secret data LQKH[DGHFLPDOIRUP (in hexadecimal form),WKH the
6')*RIEOXUILOWHUDSSOLFDWLRQ
SDFG of blur filter application SDUDOOHOFRPSXWDWLRQIRUWZR
(parallel computation for two- initial state matrix is formed as per chosen key value for state
LQLWLDOVWDWHPDWUL[LVIRUPHGDVSHUFKRVHQNH\YDOXHIRUVWDWH
pixel YDOXHV
SL[HO values 2O; DQG
and 2QO 2) VFKHGXOHG
scheduled ZLWK
with GHVLJQHUVSHFLILF
designer-specific PDWUL[JHQHUDWLRQ/HW,3YHQGRUVHOHFWHGNH\IRUVWDWHPDWUL[
matrix generation. Let IP vendor selected key for state matrix

8686

Authorized licensed use limited to: Amrita School of Engineering. Downloaded on October 04,2023 at 06:29:35 UTC from IEEE Xplore. Restrictions apply.
 generation is "01" 6HOHFWHOHPHQWVDQGVNLSWKHVXEVHTXHQW
JHQHUDWLRQLV (Select 4 elements and skip the subsequent 7TABLE
$%/(9 V
)FINAL OBTAINED DIGIT EQUIVALENTS AFTER ALPHABETIC SUBSTITUTION 
,1$/2%7$,1('',*,7(48,9$/(176$)7(5$/3+$%(7,&68%67,787,21
4).7KHREWDLQHGVWDWHPDWUL[FRUUHVSRQGLQJWRWKHVHFUHWGDWD
The obtained state matrix corresponding to the secret data
Assumed
$VVXPHG Alphabet | &RUUHVSRQGLQJ
$OSKDEHW Corresponding | 'HILQHG
Defined Output
2XWSXW
JHQHUDWHGIURP5$7RIEOXUILOWHU,3FRUHLVVKRZQLQ7DEOH,,
generated from RAT of blur filter IP core is shown in Table II. key
NH\ TRIFID
75,),' tule
UXOH
)XUWKHU
Further, 7DEOH
Table ,,, HI GHSLFWV
depicts QRQOLQHDU
non-linear ELWPDQLSXODWLRQ
bit-manipulation RQ on cipher state
FLSKHUVWDWH
JHQHUDWHG
generated LQLWLDO
initial VWDWH
state PDWUL[
matrix XVLQJ
using IRUZDUG
forward $(6
AES-128 6ER[
S-box. 100
 A
$ 331
 DE F
(atb)/e 6
Row GLIIXVLRQ
5RZ diffusion LVis SHUIRUPHG
performed EDVHGbased RQ
on WKH
the ,3
IP YHQGRU
vendor FKRVHQ
chosen 010
 %B 122
 [b+c-a 
_EFD_ 3
101
 Cc
& 222
 (atb) F*c
DE 8
key value 2. Let IP vendor selected key for row diffusion: 10
NH\YDOXH/HW,3YHQGRUVHOHFWHGNH\IRUURZGLIIXVLRQ
oll
 D
' 233
 la-b+e|
_DEF_ 2
(DFKELWRINH\YDOXHLVIRUHDFKURZRI
00 11 01 00 11 10. Each 2-bit of key value is for each row of 000
 (E 212
 D a*b*c
E F 4
VWDWH
state PDWUL[
matrix REWDLQHG
obtained DIWHU
after ELWPDQLSXODWLRQ
bit-manipulation VXFK (such DV
as 
"10" - 001
 )F 313
 atbte
DEF 7
SHUIRUP
perform FLUFXODU
circular VKLIW
shift RSHUDWLRQ
operation E\by WKUHH
three HOHPHQWV
elements DQG
and
7TABLE VI
$%/(9, 7TABLE VII
$%/(9,,
VLPLODUO\IRUUHPDLQLQJURZV
similarly for remaining rows).7KHVWDWHPDWUL[REWDLQHGDIWHU
The state matrix obtained after 67$7(0$75,;$)7(5 67$7(0$75,;$)7(5
STATE MATRIX AFTER STATE MATRIX AFTER
ALPHABETIC SUBSTITUTION 
$/3+$%(7,&68%67,787,21 PERFORMING TRANSPOSE 
3(5)250,1*75$16326(
Square matrix |
6TXDUHPDWUL[ Square matrix 2
6TXDUHPDWUL[ Square matrix 3
6TXDUHPDWUL[
(E D
' 5R W.
: Ss
6 =Z Y
< H
+ 8U 
85_ | 
23 | 
74 | 
63 85 | 
 48 | 10 |
 34 |
 76 |
 43 |
 69

)F 7T Vv
9 M
0 x
; N
1 -J ,I .K 
48 | 53 | 
39 | 
46 23 | 
 53 | 77 |
 34 |
 62 |
 74 |
 36

$ 4
Q A
$ &
Cc %B *
G 2oO /L %B 
10_ |. 
77 | 
63 | 
38 74 | 
 39 | 63 |
 32 |
 72 |
 19 |
 26


34 | 34 | 
32 | 
78 63 | 
 46 | 38 |
 78 |
 89 |
 13 |
 98

URZ
row GLIIXVLRQ
diffusion LV is VKRZQ
shown LQ in 7DEOH
Table ,9 IV. 1RZNow Da FKDUDFWHU
27-character NH\ key
76
 | 62 | 72 |
 89

has EHHQ
KDV been XVHG
used WR to SHUIRUP
perform WKH the DOSKDEHW
alphabet VXEVWLWXWLRQ
substitution XVLQJ using 43
 | 74 | 19 |
 13

75,),'
TRIFID FLSKHU
cipher LQ in 7DEOH
Table ,9 IV DV (as WKHthe WRWDO
total DOSKDEHW
alphabet LV is 26 69
 | 36 | 26 |
 98

characters ORQJ
FKDUDFWHUV long DQG
and µ¶ ‘$’ LQFOXGHG
included PDNHV makes LWit FKDUDFWHUV
27-characters). $OO All FROXPQ
column RI of 7DEOH
Table 9,,VII, NH\
key YDOXH
value LV is 
"010". 6R So, WKH
the HOHPHQWV
elements' 
FKDUDFWHUVDUHDUUDQJHGLQDWKUHHVTXDUHPDWUL[RIVL]H
27 characters are arranged in a three-square matrix of size 3*3 concatenation order per the defined rule given in Fig. 1 is &
FRQFDWHQDWLRQRUGHUSHUWKHGHILQHGUXOHJLYHQLQ)LJLV (CO,
each. 7KH
HDFK The RXWSXW
output RI of WKH
the 75,),'
TRIFID FLSKHU cipher LV is UHSUHVHQWHG
represented ZLWK with Da &
C2, &
Cl, & C3). 7KHUHIRUH
Therefore, WKH the FRQFDWHQDWHG
concatenated E\WH byte YDOXH
value IRU
for WKH
the
VWDWH
state DEF
"abc", ZKHUH
where D "a" GHQRWHV
denotes URZ row QXPEHU
number, E "b" GHQRWHV
denotes ILUVWFROXPQLV6LPLODUO\DOOUHPDLQLQJHOHPHQWV
first column is "85742363". Similarly, all remaining elements
FROXPQ
column QXPEHU
number, DQG and F "c" GHQRWHV
denotes VTXDUH square PDWUL[
matrix QXPEHU
number, of the remaining columns are concatenated as per the defined
RIWKHUHPDLQLQJFROXPQVDUHFRQFDWHQDWHGDVSHUWKHGHILQHG
UHVSHFWLYHO\IRUWKHDOSKDEHW
respectively, for the alphabet. UXOH
tule DJDLQVW
against HDFK each NH\ key YDOXH
value IRUfor E\WH
byte FRQFDWHQDWLRQ
concatenation, DQG and WKHthe

final VWULQJ
ILQDO string RI of HOHPHQWV
elements LV is JHQHUDWHG
generated. 7KH The JHQHUDWHG
generated ILQDO final
&RPSXWLQJ75,),'FLSKHURQ$
Computing TRIFID cipher on "A": sequence after byte concatenation is:
VHTXHQFHDIWHUE\WHFRQFDWHQDWLRQLV
/HWFKRVHQNH\('5)794$:6=0;1&%*<+8-,.2/3
Let chosen key: EDRFTV$QAWSZMXNCBGYHUJIKOLP "8574236348463953 10776338347834327662897243 19 13746

+HUHURZQXPEHU
Here, row number D(a)LVFROXPQQXPEHU
is 3, column number E (b)LVDQGVTXDUH
is 3, and square 9362698".

matrix F(c)QXPEHULV7KHVWDWHFRUUHVSRQGLQJWR$LV
PDWUL[ number is 1. The state corresponding to "A" is 331. Subsequently, DOO
6XEVHTXHQWO\ all GHFLPDO
decimal YDOXHV
values RI of WKH
the VWULQJ
string DUH
are FRQYHUWHG
converted
6LPLODUO\
Similarly, WKH
the VWDWH
state FRUUHVSRQGLQJ
corresponding WR to WKH
the UHPDLQLQJ
remaining DOSKDEHWV
alphabets LQWR
into WKHLU
their ELQDU\
binary HTXLYDOHQWV
equivalents, OHDGLQJleading WR to WKH
the JHQHUDWLRQ
generation RI of
is computed based on chosen key.
LVFRPSXWHGEDVHGRQFKRVHQNH\
multi-level
PXOWLOHYHO encryption
HQFU\SWLRQ and
DQG steganographic
VWHJDQRJUDSKLF security
VHFXULW\
Let ,3
/HW IP YHQGRU
vendor FKRVHQ
chosen NH\ key IRUfor DOSKDEHW
alphabet % "B": 6('5)70;
SEDRFTMX constraints-based VLJQDWXUH
FRQVWUDLQWVEDVHG signature. 7KH The JHQHUDWHG
generated ILQDO final VLJQDWXUH
signature
1&%94$:*<+2/3=8-,.
NCBVQAWGYHSOLPZUJIK. through the proposed approach is:
WKURXJKWKHSURSRVHGDSSURDFKLV
/HW
Let ,3
IP YHQGRU
vendor FKRVHQ
chosen NH\ key IRU for DOSKDEHW
alphabet & "C": 2/3=06(
OLPZMSE "1000101111100101111011100100010011011100110111101

'5)7;1&%94$:*<+8-,.
DRFTXNCBVQAWGYHUJIKS. 11111110111110001110011110001110011101111101101010

/HW
Let ,3
IP YHQGRU
vendor FKRVHQ
chosen NH\ key IRU for DOSKDEHW
alphabet ³'´ “D”’: &%9*<+
CBVGYH 00100111110100111100111111110011010011111010110100

8-,.2/3=4$:60;1('5)7
UJIKOLPZ$QAWSMXNEDRFT. 11000"

/HW
Let FKRVHQ
chosen NH\ key IRUfor DOSKDEHW
alphabet ³(´ “E”: ,.2/3=0;4$:6('5)
IKOLPZMXQAWSEDRF Further, WKH
)XUWKHU the JHQHUDWHG
generated VLJQDWXUH
signature LV is PDSSHG
mapped WR to _ LWV
its
7*<+8-1&%9
TGYHUJNCBVS. corresponding KDUGZDUH
FRUUHVSRQGLQJ hardware VHFXULW\security FRQVWUDLQWV
constraints DV as SHU
per WKH the
/HW
Let ,3
IP YHQGRU
vendor FKRVHQ
chosen NH\ key IRU for DOSKDEHW
alphabet ³)´ “F”’: %94$:6(
BVQAWSE mapping UXOH
PDSSLQJ rule RIof ,3IP YHQGRU
vendor H[SODLQHG
(explained LQ in VHFWLRQ
section ,,II(B)).
%  7KH The
'<+8-,./3=0;1&25)7*
DYHUJIKLPZMXNC$ORFTG. JHQHUDWHG
generated KDUGZDUH
hardware VHFXULW\ security FRQVWUDLQWV
constraints DUH are ==!
<Z0,Z2>,
6R
So, WKH
the REWDLQHG
obtained VWDWHstate IRUfor %
"B, &C, &
C, ' D, (
E" LVis 
"122, 
222, 
233, ==!
<Z0,Z4>, ==!==!
<Z0,Z6>,------- .<Z12,Z16>, ==! <Z12,Z18>, ==! <Z1,Z3>,
212, 
 313". 1RZ
Now, WKH the ,3IP YHQGRU
vendor VHOHFWHG
selected NH\ key IRUfor DOSKDEHWLF
alphabetic ==!==!
<Z1,Z5>,-------- »<Z7,Z19>, ==! <Z7,Z21>. 6XEVHTXHQWO\
Subsequently, DOO all WKH
the
VXEVWLWXWLRQ
substitution FRUUHVSRQGLQJ
corresponding WR to WKH
the VWDWH
state REWDLQHG
obtained DJDLQVW
against HDFK
each generated KDUGZDUH
JHQHUDWHG hardware VHFXULW\security FRQVWUDLQWV
constraints DUH are HPEHGGHG
embedded LQ in WKH
the
DOSKDEHWSUHVHQWLQWKHVWDWHPDWUL[
alphabet present in the state matrix. EOXUILOWHUDSSOLFDWLRQGHVLJQIORZXVLQJDKLJKOHYHOV\QWKHVLV
blur filter application design flow using a high-level synthesis
6R
So, OHW
let WKH
the ,3
IP YHQGRU
vendor VHOHFWHG
selected NH\ key LVis: 
"100 010 
101 
011 000 IUDPHZRUN
framework. :KLOH While HPEHGGLQJ
embedding WKH the VHFXULW\
security FRQVWUDLQWV
constraints LQ in WKH
the

001". 7DEOH
Table 9,VI UHSRUWV
reports WKH the VWDWH
state PDWUL[
matrix DIWHU
after 75,),'
TRIFID FLSKHU
cipher RAT, WZR
5$7 two VWRUDJH
storage YDULDEOH
variable SDLUV
pairs FDQQRW
cannot EH be DVVLJQHG
assigned WR to WKH
the
FRPSXWDWLRQ
computation DQG and DOSKDEHWLF
alphabetic VXEVWLWXWLRQ
substitution. )XUWKHU
Further, 7DEOHTable 9,,VII VDPH
same UHJLVWHU
register. 7KLV This LQGLFDWHV
indicates HLWKHUeither VZDSSLQJ
swapping EHWZHHQbetween
shows the transposed state matrix post-alphabetic substitution.
VKRZVWKHWUDQVSRVHGVWDWHPDWUL[SRVWDOSKDEHWLFVXEVWLWXWLRQ UHJLVWHUV
registers FRORUV
(colors)  RU or FUHDWLQJ
creating Da QHZ new UHJLVWHU
register FRORU(color)  WR to
1H[WWKHNH\YDOXHIRUE\WHFRQFDWHQDWLRQLVFKRVHQE\WKH,3
Next, the key value for byte concatenation is chosen by the IP accommodate this rule. The modified register allocation table
DFFRPPRGDWHWKLVUXOH7KHPRGLILHGUHJLVWHUDOORFDWLRQWDEOH
YHQGRU
vendor, DQG
and DOO
all HOHPHQWV
elements YDOXHV (values)  RI of WUDQVSRVHG
transposed VWDWH state PDWUL[
matrix corresponding WR
FRUUHVSRQGLQJ to PXOWLOHYHO
multi-level HQFU\SWLRQ
encryption DQG and VWHJDQRJUDSKLF
steganographic
DUH VHFXULW\
security FRQVWUDLQWV
constraints JHQHUDWHG
generated VHFXUHG
secured EOXU blur ILOWHU
filter ,3
IP FRUH
core LV is
are FRQFDWHQDWHG
concatenated DFFRUGLQJ
according WR to WKH
the UXOH
rule FRUUHVSRQGLQJ
corresponding WR to ,3
IP
depicted in Table I ZLWKFKDQJHVLQUHGFRORUHGUHJLVWHUV
GHSLFWHGLQ7DEOH, (with changes in red colored registers).7KH The
vendor chosen key value for byte concatenation. Let IP vendor
YHQGRUFKRVHQNH\YDOXHIRUE\WHFRQFDWHQDWLRQ/HW,3YHQGRU
huge YDULDWLRQ
KXJH variation LQ in WKH
the NH\
key VHOHFWLRQ
selection DW at GLIIHUHQW
different OHYHOV
levels RI of
VHOHFWHGNH\IRUE\WHFRQFDWHQDWLRQLV
selected key for byte concatenation is: "010 101 000 100 001
 encryption DQG
HQFU\SWLRQ and WKH the XVHuse RI
of VHFUHW
secret VWHJDQRJUDSKLF
steganographic GDWD data IRUfor
011 
000". 7KHUH
There ZLOO will EH be Da NH\ key YDOXHvalue IRU for HDFK
each FROXPQ
column
signature JHQHUDWLRQ
VLJQDWXUH generation LQFUHDVHV
increases WKH the UREXVWQHVV
robustness RI of WKH
the SURSRVHG
proposed
FRUUHVSRQGLQJWRWUDQVSRVHGPDWUL[)RUH[DPSOHIRUWKHILUVW
corresponding to transposed matrix. For example, for the first KDUGZDUHVHFXULW\PHWKRGRORJ\IRULPDJHILOWHUV
hardware security methodology for image filters.

87
87
Authorized licensed use limited to: Amrita School of Engineering. Downloaded on October 04,2023 at 06:29:35 UTC from IEEE Xplore. Restrictions apply.
 TABLE VIII
7$%/(9,,,
Area, Latency,\&RVWDQG5HVRXUFHFRQIL
$UHD/DWHQF Cost, and Resource configuration of Sproposed
JXUDWLRQRI hardware security\PHWKRGROR
URS RVHGKDUGZDUHVHFXULW methodologyJ\ 
Benchmarks
%HQFKPDUNV  Baseline design (before
%DVHOLQHGHVLJQ signature
EHIRUHVLJQDWXUH Signature embedded design
6LJQDWXUHHPEHGGHGGHVLJQ 
Resource
5HVRXUFH embedding) 
HPEHGGLQJ Design cost
'HVLJQFRVW
configur
FRQILJXU Design
'HVLJQ Design
'HVLJQ Design cost
'HVLJQFRVW Design area
'HVLJQDUHD Design
'HVLJQ Design cost
'HVLJQFRVW overhead
RYHUKHDG
ation
DWLRQ area(um)
DUHD latency (ps)
XP  ODWHQF\ SV  (um)
XP  latency (ps)
ODWHQF\ SV  %

Blur filter
%OXUILOWHU 1(+),
 1(*)  110.10
 1523.58
 0.673
 110.10
 1523.58
 0.673
 0

Sharpening filter
6KDUSHQLQJILOWHU 1(+),
 1(*)  111.67
 1921.04
 0.675
 111.67
 1921.04
 0.675
 0

Laplace edge detection
/DSODFHHGJHGHWHFWLRQ —_1(+),
 1(*)  105.38
 1258.61
 0.722
 105.38
 1258.61
 0.722
 0

filter
ILOWHU
TABLE IX Where ‘q’ and ‘t’ are types of encoding bits present in the
:KHUHµT¶DQGµW¶DUHW\SHVRIHQFRGLQJELWVSUHVHQWLQWKH
7$%/(,;
Pe and TT son bet d and [12] mapping UXOH
PDSSLQJ rule DQG
and VWUHQJWK
strength (size) of JHQHUDWHG
VL]H  RI generated VHFXULW\
security
3&DQG77FRPSDULVRQEHWZHHQSURSRVHGDQG>@
Benchmarks can _—— ase = pip ~ _ TT constraints
FRQVWUDLQWV respectively.
UHVSHFWLYHO\ The
7KH values
YDOXHV of 3F
RI Pc and
DQG TT
77
enenmar
%HQFKPDUNV
(Propose
3&
d)
3URSRVHG 
(Proposed)
77
3URSRVHG 
e [12]
3&>@
[12]
77>@
corresponding to blur, sharpening, and Laplace edge detection
FRUUHVSRQGLQJWREOXUVKDUSHQLQJDQG/DSODFHHGJHGHWHFWLRQ
Blur filter
%OXUILOWHU 371B-04
( 2 28E+46
( 598E-01.
( 4.19E+06
( filters DQG
ILOWHUV and LWV
its FRPSDULVRQ
comparison ZLWK
with >@
[12] DUH
are KLJKOLJKWHG
highlighted LQ
in 7DEOH
Table
Sharpening
6KDUSHQLQJ 1012E-03
( 8.92E+43
( 5.72E-01
( 1.04E+06
( Ix.
,;
filter
ILOWHU
Laplace edge
/DSODFHHGJH 3.87E-04
( 8.11E+31
( 5.52E-01
( 3.27E+04
( IV. &
,9 CONCLUSION
21&/86,21
detection
GHWHFWLRQ . . .
filter
ILOWHU A QRYHO
$ novel PXOWLOHYHO
multi-level HQFU\SWLRQ
encryption DQG
and VWHJDQRJUDSKLF
steganographic
security constraints-based KDUGZDUHVHFXULW\
VHFXULW\FRQVWUDLQWVEDVHG hardware security PHWKRGRORJ\
methodology IRUfor
,,,
III. 5RESULTS AND '
(68/76$1' ,6&866,21
DISCUSSION securing image processing IP cores has been presented in the
VHFXULQJLPDJHSURFHVVLQJ,3FRUHVKDVEHHQSUHVHQWHGLQWKH
paper. 7KH
SDSHU The SURSRVHG
proposed DSSURDFK
approach RIIHUV
offers PRUH
more UREXVW
robust VHFXULW\
security
The SURSRVHG
7KH proposed KDUGZDUH
hardware VHFXULW\
security PHWKRGRORJ\
methodology LV
is DQDO\]HG
analyzed against FRXQWHUIHLWLQJ
DJDLQVW counterfeiting, SLUDF\
piracy, DQG
and IUDXGXOHQW
fraudulent FODLP
claim RIof ,3
IP
regarding its security analysis and design cost overhead.
UHJDUGLQJLWVVHFXULW\DQDO\VLVDQGGHVLJQFRVWRYHUKHDG ownership ZLWK
RZQHUVKLS with WKH
the KHOS
help RIof DQ
an HPEHGGHG
embedded VLJQDWXUH
signature LQ
in WKH
the
A. Design cost Analysis
$ 'HVLJQFRVW$QDO\VLV design IORZ
GHVLJQ flow RI
of WKH
the ,3
IP FRUH
core. 7KH
The SURSRVHG
proposed DSSURDFK
approach SURYLGHV
provides
stronger VHFXULW\
VWURQJHU security ZLWK
with ]HUR
zero GHVLJQ
design FRVW
cost RYHUKHDG
overhead, DV
as HYLGHQW
evident
The used design cost metric used in the proposed approach
7KHXVHGGHVLJQFRVWPHWULFXVHGLQWKHSURSRVHGDSSURDFK
from the values of Pc and TT respectively.
IURPWKHYDOXHVRI3FDQG77UHVSHFWLYHO\
is a function of total area and latency:
LVDIXQFWLRQRIWRWDODUHDDQGODWHQF\
୅୰ୣୟ
Area ୐ୟ୲ୣ୬ୡ୷
Latency REFERENCES
Cost
‘•– =ൌ tl
–ͳ ‫——— *כ‬ +൅t2–ʹ *‫כ‬    5()(5(1&(6
୑ୟ୶ୟ୰ୣୟ
Max area ୑ୟ୶୧୫୳୫୪ୟ୲ୣ୬ୡ୷
Maximum latency M
Where µƒ”‡ƒ¶
:KHUH ‘area’ DQG
and µODWHQF\¶
‘latency’ UHSUHVHQWV
represents WKHthe WRWDO
total DUHD
area DQG
and (1) 5
>@ R. Schneiderman,
6FKQHLGHUPDQ “DSPs
³'63V evolving
HYROYLQJ in FRQVXPHU
LQ consumer electronics
HOHFWURQLFV
applications,” JEEE Signal Process. Mag., vol. 27 (3),
DSSOLFDWLRQV´,(((6LJQDO3URFHVV0DJYRO pp. 6-10, 2010.
 SS±
latency (delay)
ODWHQF\ of the proposed methodology-based secured IP
GHOD\ RIWKHSURSRVHGPHWKRGRORJ\EDVHGVHFXUHG,3
[2] )
>@ F. .RXVKDQIDU
Koushanfar HW et DO
al., ³&DQ
“Can ('$ EDA FRPEDWcombat WKH the ULVH
rise RI of HOHFWURQLF
electronic
core design; ‘max area and max latency’ depict the maximum
FRUHGHVLJQµPD[DUHDDQGPD[ODWHQF\¶GHSLFWWKHPD[LPXP counterfeiting?,” JEEE DAC, CA, 2012, pp. 133-138.
FRXQWHUIHLWLQJ"´,((('$&&$SS
area DQG
DUHD and ODWHQF\
latency RIof WKH
the SURSRVHG
proposed VHFXUHG
secured GHVLJQ
design RIof ,3
IP FRUH
core [3] 6
>@ S. 6DGDQJL
Sadangi, 6 S. %DUDKD
Baraha, ' D. . K. 6DWSDWK\
Satpathy, DQG and 3P. .
K. %LVZDO
Biswal, ³)3*$ “FPGA
using PD[LPXP
XVLQJ maximum UHVRXUFH
resource FRQVWUDLQWV
constraints SRVVLEOH
possible. µW
‘tl DQG
and W¶
t2’ DUH
are implementation RI
LPSOHPHQWDWLRQ of VSDWLDO
spatial ILOWHULQJ
filtering WHFKQLTXHV
techniques IRU for '
2D LPDJHV´
images,” LQ in 3URF
Proc.
the ZHLJKLQJ
WKH weighing IDFWRUV
factors (weightage given WR
ZHLJKWDJH JLYHQ to DUH
are DQG
and GHOD\
delay),  RTEICT, 2017, pp. 1213-1217.
57(,&7SS±
which LQ
ZKLFK in WKH
the SURSRVHG
proposed DSSURDFK
approach LVis 
0.5 HDFK
each. 7KH
The SURSRVHG
proposed [4] '
>@ D. 7VLNWVLULV
Tsiktsiris, '
D. =LRX]LRV
Ziouzios, DQG and 0M. 'DV\JHQLV
Dasygenis, ³$ “A SRUWDEOH
portable LPDJH image
processing accelerator using FPGA,” in Proc. MOCAST, 2018, pp. 14.
SURFHVVLQJDFFHOHUDWRUXVLQJ)3*$´LQ3URF02&$67SS±
approach offers a zero overhead in terms of design cost before
DSSURDFKRIIHUVD]HURRYHUKHDGLQWHUPVRIGHVLJQFRVWEHIRUH
and
DQG after
DIWHU implanting
LPSODQWLQJ proposed
SURSRVHG hardware _ VHFXULW\
KDUGZDUH security [5] -
>@ J. 9
V. 9RXUYRXODNLV
Vourvoulakis, - J. /\JRXUDV
Lygouras, DQG and -
J. $
A. .DORPLURV
Kalomiros, ³DFFHOHUDWLRQ
“acceleration RI of
image SURFHVVLQJ
LPDJH processing DOJRULWKPV
algorithms XVLQJ using PLQLPDO
minimal UHVRXUFHV
resources RIFXVWRP
ofcustom
methodology-based security constraints. Table VIII shows the
PHWKRGRORJ\EDVHGVHFXULW\FRQVWUDLQWV7DEOH9,,,VKRZVWKH reconfigurable hardware,” in Proc. PCI, 2012, pp. 68-73.
UHFRQILJXUDEOHKDUGZDUH´LQ3URF3&,SS±
design cost corresponding to image processing filter IP cores.
GHVLJQFRVWFRUUHVSRQGLQJWRLPDJHSURFHVVLQJILOWHU,3FRUHV [6]
>@ D. 'DYDOOH
' Davalle, % B. &DUQHYDOH
Carnevale, 6 S. 6DSRQDUD
Saponara, / L. )DQXFFL
Fanucci, DQGand 3 P. 7HUUHQL
Terreni,
“Hardware DFFHOHUDWRU
³+DUGZDUH accelerator IRU for IDVW
fast LPDJHYLGHR
image/video WKLQQLQJ´
thinning,” LQ in 3URF
Proc. ,67IST,
2014, pp. 64-67,.
SS±
B. 6HFXULW\$QDO\VLV
% Security Analysis [7] A. Ansari, K. Gunnam, and T. Ogunfunmi, “An efficient reconfigurable
>@$$QVDUL.*XQQDPDQG72JXQIXQPL³$QHIILFLHQWUHFRQILJXUDEOH
hardware DFFHOHUDWRU
KDUGZDUH accelerator IRU for FRQYROXWLRQDO
convolutional QHXUDO neural QHWZRUNV´
networks,” LQ in 3URF
Proc.
Probability of coincidence (Pc)
3UREDELOLW\RIFRLQFLGHQFH and tamper tolerance (TT)
3F DQGWDPSHUWROHUDQFH 77 
ACSSC, Pacific Grove, CA, USA, 2017, pp. 1337-1341.
$&66&3DFLILF*URYH&$86$SS±
are the two PDMRU
DUHWKHWZR major PHWULFVEDVHGRQ
metrics based on ZKLFKWKHVHFXULW\RI
which the security of WKHthe
hardware security technique is analyzed. The lower the Pc, the
KDUGZDUHVHFXULW\WHFKQLTXHLVDQDO\]HG7KHORZHUWKH3FWKH [8]   &
>@ C. 6KX
Shu, :
W. 3DQJ
Pang, + H. /LX
Liu, DQGand 6 S. /X
Lu, ³+LJK
“High HQHUJ\
energy HIILFLHQF\
efficiency )3*$ FPGA
based DFFHOHUDWRU
EDVHG accelerator IRU for FRQYROXWLRQDO
convolutional QHXUDO neural QHWZRUNV
networks XVLQJ using ZHLJKW
weight
stronger LV
VWURQJHU is WKH
the VHFXULW\
security RI
of WKH
the SURSRVHG
proposed DSSURDFK
approach LQ
in WHUPV
terms RI
of combination,” in Proc. ICSIP, Wuxi, China, 2019, pp. 578-582.
FRPELQDWLRQ´LQ3URF,&6,3:X[L&KLQDSS±
secret digital evidence. The higher the TT, the more robust is
VHFUHWGLJLWDOHYLGHQFH7KHKLJKHUWKH77WKHPRUHUREXVWLV [9]  6
>@ S. %
B. &KRL
Choi, 6
S. 6
S. /HH
Lee, DQGand 6 S. -
J. -DQJ
Jang, ³&11
“CNN LQIHUHQFH
inference VLPXODWRU
simulator IRU for
the SURSRVHG
WKH proposed DSSURDFK
approach DJDLQVW
against WDPSHULQJ
tampering DWWDFN
attack. 7KHUHIRUH
Therefore, accurate DQG
DFFXUDWH and HIILFLHQWDFFHOHUDWRU
efficient accelerator GHVLJQ´ design,” LQ in 3URF
Proc. ,62&&
ISOCC, -HMXJeju, 6RXWK
South
tampering DQ
WDPSHULQJ an ,3
IP FRUH
core GHVLJQ
design IRU
for HYDGLQJ
evading WKH
the FRXQWHUIHLW
counterfeit Korea, 2019, pp. 283-284.
.RUHDSS±
detection SURFHVV
GHWHFWLRQ process EHFRPHV
becomes FRPSOH[
complex, ZLWK
with D
a KLJKHU
higher 77
TT YDOXH
value [10] 0
>@ M. &KDQJ
Chang, =Z. 3DQ
Pan, DQG and -J. &KHQ
Chen, ³+DUGZDUH
“Hardware DFFHOHUDWRU
accelerator IRU for ERRVWLQJ
boosting
for an adversary. The Pc metric is expressed as:
IRUDQDGYHUVDU\7KH3FPHWULFLVH[SUHVVHGDV convolution FRPSXWDWLRQ
FRQYROXWLRQ computation LQ in LPDJH
image FODVVLILFDWLRQ
classification DSSOLFDWLRQV´
applications,” LQ in 3URF
Proc.
ଵ ௭zZ GCCE, 2017.
*&&(
Pc ൌ = ቀͳ
(1 െ- ௫=)ቁ  (5)  [11] W. Zamir, A. Arora, S. Khan, F. S. Khan, and L. Shao, Learning Digital
>@:=DPLU$$URUD6.KDQ)6.KDQDQG/6KDR/HDUQLQJ'LJLWDO
Where 'x'
:KHUH and 'z’
[  DQG are WKH
]¶ DUH the QXPEHUV
numbers RI of UHJLVWHUV
registers UHTXLUHG
required Camera Pipeline for Extreme Low-Light Imaging. Accessed: Jun. 2020.
&DPHUD3LSHOLQHIRU([WUHPH/RZ/LJKW,PDJLQJ$FFHVVHG-XQ

before encrypted stego-constraints embedding and the strength
EHIRUHHQFU\SWHGVWHJRFRQVWUDLQWVHPEHGGLQJDQGWKHVWUHQJWK
[12] )
>@ F. .RXVKDQIDU
Koushanfar, , I. +RQJ
Hong, DQG and 0 M. 3RWNRQMDN
Potkonjak, %HKDYLRUDO
"Behavioral V\QWKHVLV
synthesis
of LPSODQWHG
RI implanted VHFXULW\
security FRQVWUDLQWV
constraints.  )XUWKHU
Further, WKH the 77
TT PHWULF
metric LV is techniques IRU
WHFKQLTXHV for LQWHOOHFWXDO
intellectual SURSHUW\
property SURWHFWLRQ
protection," $&0 ACM 7UDQV
Trans. 'HVLJQ
Design
expressed as:
H[SUHVVHGDV Autom. (OHFWURQ
$XWRP Electron. 6\VW
Syst., YROvol. 10, QR
no.  3, SS
pp. ±
523-545, -XO Jul. 
2005.
 ൌ ‫ ݍ‬௧  (6)
TT=q' 

88
88
Authorized licensed use limited to: Amrita School of Engineering. Downloaded on October 04,2023 at 06:29:35 UTC from IEEE Xplore. Restrictions apply.