115
“Al-Fa g‘oniy a lodla i” elek on ilmiy
ju nali
ISSN 2181-4252. Tom: 1 | Son: 3 | 2025-yil
"Descendan s o Al-Fa ghani" elec onic scien i ic
jou nal.
ISSN 2181-4252. Vol: 1 | Iss: 3 | 2025 yea
Электронный научный журнал "Потомки Аль-
Фаргани"
ISSN 2181-4252. Том: 1 | Выпуск: 3 | 2025 год
h ps://al- a goniy.uz/
DETECTION OF PRINTED FALSE ATTACKS USING NEURAL NETWORKS
Abdukadi o Bakh iyo ,
Fe gana S a e Uni e si y
Docen , Depa men o In o ma ion Technology
e-mail: [email p o ec ed]
Abs ac . This a icle discusses a me hod o de ec ing alse posi i es agains a biome ic acial
ecogni ion sys em based on deep con olu ional neu al ne wo ks. The p oposed me hod is designed
o de ec p in ed alse posi i es and is es ed on open da abases o eal and ake aces, and he esul s
a e analyzed. The ypes o alse posi i e a acks launched agains a biome ic sys em based on
exis ing aces a e analyzed.
Keywo ds: biome ic sys em, alse ala m a ack, local bina y pa e n, suppo ec o machine,
con olu ional neu al ne wo ks, ecu en ne wo k, classi ica ion e alua ion me ics.
In oduc ion. Wi h he widesp ead adop ion
and inc eased e ec i eness o biome ic ecogni ion
sys ems, he numbe o a emp s by po en ial a acke s
o di ec ly hack he sys em, o example, by logging in
as a egula use , has inc eased. Such a emp s a e
called spoo ing a acks [1]. Mos exis ing biome ic
sys ems a e ulne able o spoo ing a acks. Fo
biome ic sys ems, a spoo ing a ack in ol es
decei ing he sys em by p esen ing pho og aphs,
ideos, o p e- eco ded sounds o he biome ic senso ,
wi h he goal o impe sona ing an un egis e ed use
du ing he ecogni ion p ocess. [2] success ully
demons a ed hacking comme cial acial ecogni ion
sys ems using a pho og aph o ideo eco ding o a
egis e ed use using he de ice's sc een.
Acco ding o he de elope s, he bes esul s
we e demons a ed by sys ems using specialized
scanne s o ideo came as ha allow 3D objec
econs uc ion. Howe e , me hods o con i ming he
au hen ici y o a ecognized objec ha do no equi e
specialized equipmen and do no equi e addi ional
ac ions a e he mos p omising, as hey a e mo e
con enien o he end use and can be easily in eg a ed
in o exis ing acial ecogni ion sys ems.
An addi ional challenge o de elope s o an i-
spoo ing sys ems is he lack o public da abases
con aining a comple e lis o spoo ing a acks (pape
pho og aphs o he objec , pho os and ideos om
a ious de ice sc eens, pho o masks, silicone masks,
images o he objec wi h applied makeup, as well as
3D masks and dummies). Many esea che s es
de eloped objec au hen ica ion solu ions on hei own
da abases, which a e no publicly a ailable. The lack o
publicly a ailable da abases hinde s a ai e alua ion
and compa ison o p oposed me hods o con i ming
he au hen ici y o a ecognized objec .
Li e a u e analysis and me hodology. The e
a e a numbe o s udies ha use a ious ex u al
ea u es o de ec spoo ing: LBP and suppo ec o
machines ( he pe cen age o co ec ecogni ion o a
eal objec was a leas 91.2%, while he pe cen age o
a acke s missed was no mo e han 0.2%) [3, 4], LBP
and a i icial neu al ne wo ks ( he pe cen age o co ec
ecogni ion was 97.5% based on wo-dimensional
images [5], Gabo wa ele s [4], his og ams o o ien ed
g adien s [6], e c.
In [7], images o he use 's pupils a e used o
con i m he au hen ici y o he ecognized objec . In
his in en ion, an i-spoo ing p o ec ion is based on he
p ope y o he human pupil o cons ic wi h inc easing
ligh in ensi y. To de e mine whe he he use 's image
was ob ained om he eye o a li ing pe son and no
om a dummy, he illumina ion in ensi y is a ied. The
an i-spoo ing sys em moni o s he pupil's esponse o
ligh modula ion, he e o e sligh ly inc easing he
egis a ion ime.
In [8], he use is asked o look a andomly
assigned loca ions on he moni o . Eye mo emen
ajec o ies a e hen analyzed. The me hod's de elope s
epo a 95% co ec ecogni ion a e o subs i u ions
116
“Al-Fa g‘oniy a lodla i” elek on ilmiy
ju nali
ISSN 2181-4252. Tom: 1 | Son: 3 | 2025-yil
"Descendan s o Al-Fa ghani" elec onic scien i ic
jou nal.
ISSN 2181-4252. Vol: 1 | Iss: 3 | 2025 yea
Электронный научный журнал "Потомки Аль-
Фаргани"
ISSN 2181-4252. Том: 1 | Выпуск: 3 | 2025 год
h ps://al- a goniy.uz/
in a da abase con aining wo-dimensional images o he
objec being ecognized.
A combina ion o di e en app oaches is also
possible o achie e highe esul s. In [9], he use is
asked o make a mo emen speci ied by he sys em,
change acial exp essions, open hei mou h, e c. The
sequence o images om an IR came a is hen analyzed
a e no maliza ion and noise emo al. A combina ion
o se e al app oaches is also p esen ed in [10]. The
p esen ed objec subs i u ion de ec ion sys em uses a
ideo came a, mo ion senso s, and ligh senso s. The
esea che s ob ained he ollowing esul s: ecogni ion
speed was 3 seconds, he pe cen age o co ec
ecogni ion o he eal objec was 95-97%, and he
pe cen age o in ude s missed was 2-3%. The
disad an ages o such me hods include he need o
addi ional equipmen o incon enience o he end use
(an un iendly in e ace o he au hen ica ion sys em).
Cu en ly, he e a e no echnologies capable o
p o iding eliable ecogni ion wi hou specialized
equipmen . To ensu e eliable ecogni ion, i is
necessa y o ensu e a high p obabili y o co ec acial
image ecogni ion, con i m he au hen ici y o he
ecognized objec , and enable he ecogni ion sys em o
ope a e in eal ime. Exis ing echnologies can only
pe o m hese ope a ions indi idually. The e o e, i has
become necessa y o de elop app op ia e echnology
and indi idual componen s o he ecogni ion p ocess.
Ve i ica ion sys ems a e p ima ily ocused on
ecognizing he use 's ace and do no p o ec agains
subs i u ion o he ecognized objec . In [2], he au ho s
demons a ed a success ul hack o comme cial acial
ecogni ion sys ems using a pho og aph o ideo
eco ding o a egis e ed use . Biome ic ecogni ion
sys ems equi e he implemen a ion o e ec i e
me hods o con i ming he au hen ici y o he
ecognized objec .
A ack Model o Biome ic Sys ems. When
de eloping secu i y me hods o biome ic use
ecogni ion sys ems, i is necessa y o iden i y all
possible h ea s and desc ibe he a ack model. A ack
models o a ious biome ic sys ems ha e al eady
been de eloped [11, 12]. This sec ion desc ibes an
a ack model o acial ecogni ion sys ems. Nine o he
mos ulne able a eas o a ack by a acke s a e
iden i ied in he gene al biome ic sys em diag am
shown in Figu e 1.
Fig. 1. Gene al diag am o a biome ic use
ecogni ion sys em wi h a acks iden i ied
To ca y ou success ul a acks on biome ic
ecogni ion sys ems, an a acke mus ha e skills in
a ious specialized ields, as well as ha e in o ma ion
abou laws in he equipmen and ha dwa e
implemen a ion, he s uc u e and me hod o
o ganizing he da abase, me hods o calcula ing and
compa ing in o ma i e ea u es, abou he subsys em
o in e ac ing wi h he da abase, and abou o he
models and me hods embedded in he implemen ed
biome ic ecogni ion sys em [11].
Iden i ying alse a acks. Unlike sys ems ha
equi e addi ional equipmen ( inge p in ecogni ion,
e c.), ecognizing a pe son om a acial image makes
i e y easy o c ea e a eplica o he a ge objec . All
ha is needed is a pho og aph o he pe son, which can
be easily ound online o pho og aphed emo ely. One
o he objec i es o he s udy is o de ec he
subs i u ion o an objec p esen ed o a ideo came a.
This ask is no always easy, e en o humans. The
solu ion mus ha e low compu a ional complexi y and
a high p obabili y o co ec ly de ec ing he
subs i u ion o a ecognized objec wi hou he use o
addi ional specialized equipmen . Depending on he
ocus o he biome ic sys em, spoo ing a acks can
ha e a ying le els o complexi y.
In ecen yea s, deep lea ning me hods ha e
been ac i ely used o sol e his p oblem. In [20], an
AlexNe - ype CNN was used o bina y classi ica ion
117
“Al-Fa g‘oniy a lodla i” elek on ilmiy
ju nali
ISSN 2181-4252. Tom: 1 | Son: 3 | 2025-yil
"Descendan s o Al-Fa ghani" elec onic scien i ic
jou nal.
ISSN 2181-4252. Vol: 1 | Iss: 3 | 2025 yea
Электронный научный журнал "Потомки Аль-
Фаргани"
ISSN 2181-4252. Том: 1 | Выпуск: 3 | 2025 год
h ps://al- a goniy.uz/
o a eal ace e sus a spoo ing a ack, and he e ec o
he size o he con ex cap u ed du ing acial image
alignmen on he quali y o his classi ica ion was also
in es iga ed. The au ho s o [9] used SVM o classi y
high-le el ea u es ex ac ed om he inal laye s o a
e ained VGG-Face ne wo k. In [19], a combina ion o
a CNN and an LSTM- ype ecu en ne wo k was used
o classi y a ame sequence.
Despi e signi ican p og ess in acial i ali y
de ec ion, his p oblem emains unsol ed in gene al.
The ul a-high pixel densi y and na u al colo endi ion
o mode n displays make he displayed acial image
almos indis inguishable om he eal one. The e o e,
he e is a need o imp o e he p oposed spoo ing
de ec ion algo i hms and de elop new app oaches.
Cu en ly, deep con olu ional neu al ne wo ks
a e a s anda d building block in i ually all image
p ocessing asks. Mode n machine lea ning lib a ies
signi ican ly accele a e he de elopmen o new neu al
ne wo k a chi ec u es, which leads o a g adual
inc ease in he complexi y o hei compu a ional
g aph. Recen wo k in he ield o deep lea ning has
demons a ed he e ec i eness o he a en ion
mechanism in imp o ing he pe o mance o CNNs in
asks such as pa e n ecogni ion, image cap ion
gene a ion, and o he s [14]. Con olu ional neu al
ne wo ks a e ypically ained using he RGB channels
o an image di ec ly. Howe e , nume ous p e ious
s udies on acial i ali y de ec ion ha e demons a ed
he e ec i eness o using a ious ex u e desc ip o s o
encode he acial egion. The need o gene alize
di e ences in illumina ion and acial pose ac oss wo
isually simila classes, while simul aneously
de ec ing ine-g ained ex u e di e ences, complica es
he ask o neu al ne wo k op imiza ion o his
p oblem.
Resea ch esul s. To de ec ake a acks on
biome ic sys ems, we ain a deep neu al ne wo k o
c ea e a li eness de ec o capable o dis inguishing
be ween eal and ake aces. We conside ace li eness
de ec ion as a bina y classi ica ion p oblem.
We ain he neu al ne wo k on he NUAA
Impos e Da abase and LCC FASD da abases, which
con ain bo h eal and ake ace images. The NUAA
Impos e Da abase con ains 5,105 eal and 7,509 ake
ace images, while he LCC FASD da abase con ains
7,047 eal and 7,076 ake ace images.
To expand he aining se o p o ec agains
ake a acks, a se ies o augmen a ion p ocesses a e
used o simula e a ious e ec s, including: image
o a ion by a speci ic angle, image enla gemen , image
d agging along he wid h and heigh , and image
o a ion a ound he ho izon al axis. The aining and
es se s o all da abases a e spli 75% o 25%. The
accu acy and e o g aphs o he coun e ei a ack
de ec ion model a e aining on he abo e da abase
a e shown in Figu es 2–3.
Fig. 2. Accu acy and loss on he NUAA PI DB
da abase
Fi g. 3. Accu acy and loss on he LCC FASD da abase
The di e ence be ween ain loss and alida ion
loss is ha he o me e e s o he aining se , and he
la e e e s o he es se . Thus, he alida ion loss
shows how he model pe o ms on ex aneous da a
118
“Al-Fa g‘oniy a lodla i” elek on ilmiy
ju nali
ISSN 2181-4252. Tom: 1 | Son: 3 | 2025-yil
"Descendan s o Al-Fa ghani" elec onic scien i ic
jou nal.
ISSN 2181-4252. Vol: 1 | Iss: 3 | 2025 yea
Электронный научный журнал "Потомки Аль-
Фаргани"
ISSN 2181-4252. Том: 1 | Выпуск: 3 | 2025 год
h ps://al- a goniy.uz/
(i.e., da a no ained on his da a), and he e o e se es
as a key indica o o e alua ing he model. The
alida ion e o is also used o p e en o e aining.
The accu acy, p ecision, ecall, and 1-sco e
e alua ion me ics we e used o classi y he da a. The
esul s o compa ing he pe o mance o he h ee ypes
o da abases wi h he e alua ion a e p esen ed in Table
1.
Table 1. Pe o mance compa ison esul s
Name o
he
da abase
Pe o mance e alua ion
p ecision,
%
ecall,
%
Accu acy,
%
1, %
NUAA PI
DB
100
100
100
100
LCC FASD
95
94
94
94
The esul s p esen ed in Table 1 show ha
when he neu al ne wo k model was ained on he
NUAA PI DB da abase, o e i ing occu ed, in which
he model simply memo ized some o he da a om he
aining sample. This can also be seen om he
accu acy and e o g aph p esen ed in Figu e 2.
Al hough he esul s on o he da abases a e be e han
he i s case, hei esul s canno be called pe ec .
Conclusion. In his a icle, he p ima y ype o
decep i e a ack on biome ic iden i ica ion sys ems is
he inge p in a ack. This a ack was de ec ed by
aining a con olu ional neu al ne wo k on wo
da abases o eal and ake aces. O he decep i e
a acks lis ed abo e, he mos common a e inge p in -
based and ideo-based a acks. The esul ing li e ace
de ec ion de ec o , capable o dis inguishing be ween
eal and ake aces, can de ec no only inge p in -
based a acks bu also ideo-based a acks, bu
pe o ms poo ly in de ec ing disguised a acks using
3D aces.
The da abase also plays an impo an ole in he
de elopmen o he ake ace de ec ion de ec o . The
a o emen ioned da abases, due o hei open sou ce
a ailabili y, allow o ex ensi e expe imen s.
Howe e , mode n and p op ie a y da abases consis ing
o eal and ake ace images, such as IDIAP Replay-
A ack, PHOTO-ATTACK, CASIA FASD, MSU
Mobile Spoo ing Da abase, P in A ack, Replay
A ack, Ga ed Recu en Uni , and OULU, we e no
used. The da abases used in he expe imen s lack e hnic
di e si y, and mul imodal me hods should be used o
imp o e he e ec i eness o ake ace de ec o s.
REFERENCES
1. Фозилов Ш.Х., Раджабов С.С, Абдукадиров Б.А. /
Шахсни биометрик идентификациялаш тизимларида сохта
киришни аниқлаш муаммоси // Мухаммад ал-Хоразмий
авлодлари, Ташкент 2020. — №3(13). - Б. 16–23.
2. Duc N.M. You ace is no you passwo d / Black Ha
Con e ence. 2009. – C. 1-16.
3. Chingo ska, I. On he E ec i eness o Local Bina y
Pa e ns in Face An ispoo ing / I. Chingo ska, A. Anjos, S.
Ma cel // Biome ics Special In e es G oup, 2012 BIOSIG -
P oceedings o he In e na ional Con e ence o he.– 2012.
4. Maa a, J. Face spoo ing de ec ion om single images
using mic o- ex u e analysis / J. Maa a, A. Hadid, M.Pie ikainen
// Biome ics (IJCB), In e na ional Join Con e ence on
Biome ics, IEEE. – 2011. – С. 1-7.
5. Rekha, P.S. Spoo ing Face Recogni ion Using Neu al
Ne wo k wi h 3D Mask / / IJETCSE. – 2015. – Т. 14 – № 1 – С.
123-127.
6. Yang, J. Face li eness de ec ion wi h componen
dependen desc ip o / J. Yang, Z. Lei, S. Liao // Biome ics
(ICB). – 2013. – С. 1-6.
7. Способ идентификации личности по радужной
оболочке глаза (варианты) : пат. 2407435 Рос. Федерация :
МПК A 61 B 3/10 / Д. Е. Антонов ; заявитель и
патентообладатель Антонов Дмитрий Евгеньевич. – №
2009128069/14 ; заявл. 22.07.2009 ; опубл. 27.12.2010.
8. Adamiak, K. Li eness de ec ion in emo e biome ics
based on gaze di ec ion es ima ion / K. Adamiak, D. Zu ek, K.
Slo // P oc. Fed. Con . Compu . Sci. In . Sys . – 2015. – C. 225–
230.
9. Kan , C. Fake Face Recogni ion Using Fusion o
The mal Imaging and Skin Elas ici y / C. Kan , N. Sha ma //
IJCSC. – 2013. – Т. 4 – C. 65-72.
10. Chen, S. Senso -assis ed acial ecogni ion: an enhanced
biome ic au hen ica ion sys em o sma phones / S. Chen, A.
Pande, P. Mohapa a // MobiSys '14. – 2014. – C. 109-122.
11. Ручай, А.Н. Модель атак и защиты биометрических
систем распознавания диктора / А. Н. Ручай // Доклады
ТУСУРа. – 2011. – № 1 – С. 96–100.
12. Ra ha, N.K. Enhancing secu i y and p i acy in
biome ics-based au hen ica ion sys ems / N. K. Ra ha, J. H.
Connell, R. M. Bolle // IBM Sys . J. – 2001. – Т. 40, № 3 – С.
614–634.
13. Yang, J. Lea n con olu ional neu al ne wo k o ace
an ispoo ing. a Xi p ep in a Xi :1408.5601, 2014
14. Li, L. An o iginal ace an i-spoo ing app oach using
pa ial con olu ional neu al ne wo k. In IPTA, 2016.
15. Xu, Z. Lea ning empo al ea u es using LSTM-CNN
a chi ec u e o ace an i-spoo ing. ACPR, 2015, pp. 41-45
16. Rod iguez P., A painless a en ion mechanism o
con olu ional neu al ne wo ks, ICLR, 2018, pp. 1-8