US 20130097697A1

(19) United States

(12) Patent Application Publication (10) Pub. No.: US 2013/0097697 A1
(43) Pub. Date:

Zhu et al.
(54)

SECURITY PRIMITIVES EMPLOYING HARD

(52)

ARTIFICIAL INTELLIGENCE PROBLEMS

(75) Inventors: Bin Benjamin Zhu, Edina, MN (US);

Guanbo Bao, Beijing (CN); Maowei

Yang, Chengdu (CN); Ning Xu, Beijing

(CN)
(73) Assignee: Microsoft Corporation, Redmond, WA

(Us)
Oct. 14, 2011
Publication Classi?cation

(51)

Int. Cl.
G06F 21/00

US. Cl.
USPC

.......................................................... ..

(57)

726/18

ABSTRACT

A security module generates a random image having a plu

rality of password-element indicators therein. The random
image is provided to a user. The user selects portions of the

random image. The security module determines whether the

selected portions of the random image correspond to a pass
word for the user. The security module grants access if the

(21) Appl. No.: 13/273,841

(22) Filed:

Apr. 18, 2013

(2006.01)

selected portions of the random image correspond to the

users password. However, if the selected portions of the
random image do not correspond to the users password, the
security module may generate another random image having
a plurality of password-element indicators therein, wherein
each of the random images are computationally de-corre
lated.

100
106

>\

V
SECURITY
PRIMITIVE
DATABASE

USER
PROFILE
DATABASE
124b

108

Patent Application Publication

Apr. 18, 2013 Sheet 1 0f 9

US 2013/0097697 A1

100

SECURITY
PRIMITIVE
DATABASE

PROF I LE
DATABASE

214
206

SECURITY
MODULE
USER
PROFILES
SECURITY
PRIMITIVES

302

310 x
I.

SECURITY
I

316

MODULE

306

USER
'

/- 314

PROFILES

/- 312

SECURITY /
PRIMITIVES

FIG

304

Patent Application Publication

Apr. 18, 2013 Sheet 2 0f9

US 2013/0097697 A1

CARP ENVIRONMENT 400

PASSWORD m

PRIMITIVE DOMAIN @
PPE 1

m
PPE 4

PPE 2

PPE 3

PPE 5

PPE 6

"

f\

USER INPUT m

us 4 (x4, Y4)
418

DISPLAYED IMAGE m

414d

FIG. 4

Patent Application Publication

Apr. 18, 2013 Sheet 3 0f 9

US 2013/0097697 A1

PASSWORD 502a
PASSWORD ELEMENT
O

PASSWORD ELEMENT
504-N

PASSWORD ELEMENT

Q
FIG. 5A
PASSWORD 502b
PASSWORD ELEMEN
50

PASSWORD ELEMENT PASSWORP ELEMENT

5

SECURITY

CIIEZNT

ADMINISTRATOR

/ 806

ACCESS REQUEST (USER ID)

/

GENERATE
IMAGE

810

f- 808

PROVIDE IMAGE
/_ 812

USER INPUT [(x1, Y1), (x2, Y2) (XN, YN)]

/ 816

GRANT/DENY ACCESS

FIG. 8

_ JI

f 814

Patent Application Publication

(m

Apr. 18, 2013 Sheet 4 0f 9

US 2013/0097697 A1

USER-INTERFACE
600

FIG. 6
DEFINE PRIMITIVE DOMAIN

E
II

REGISTER USER (USER ID)

M
II

PROVIDE PASSWORD-ELEMENTS

E
II

RECEIVE USER SELECTION

GENERATE PASSWORD INDICATOR

;
ASSOCIATE PASSWORD/PASSWORD
INDICATOR WITH USER

FIG. 7

/ 700

Patent Application Publication

Apr. 18, 2013 Sheet 5 0f 9

US 2013/0097697 A1

/ 900
/
906-1 -/

\ \ 906-3

904 /
\

906-5

906-4

\ 906-2

FIG. 9

Patent Application Publication

Apr. 18, 2013 Sheet 6 0f 9

wo ?

r
Lwoo

rAmwi/?$o01z5 6529:83; N$65:803

.. '

UO

US 2013/0097697 A1

Patent Application Publication

Apr. 18, 2013 Sheet 7 0f 9

US 2013/0097697 A1

DETERMINE POTENTIAL PASSWORD ELEMENTS

1102

I
MODEL POTENTIAL PASSWORD ELEMENTS
1104

I
ARRANGE PASSWORD ELEMENT INDICATORS
1106

I
GENERATE IMAGE
1108

FIG. 11

Patent Application Publication

Apr. 18, 2013 Sheet 8 0f 9

ff

US 2013/0097697 A1

Q
1202

FIG. 12A
m

02
J)

(f
\\

1202

1206

FIG. 12B

\\
J)

(K
9

1202

(1,1) (2,1) (31)

\(13) (2,3) (3,3)

FIG 12C

Patent Application Publication

Apr. 18, 2013 Sheet 9 0f 9

US 2013/0097697 A1

1300

1302a

1304

13 20

1302b

1302d

A
1304a

FIG. 13A
f- 1414

1304b

FIG. 13B

1%

1400

/ 1402

COMPUTING DEVICE

1*

I SYSTEM MEMORY

REMOVABLE

STORAGE

1406'

ROM/RAM

1416
f

lr
/_1418
: NON REMOVABLE /

T OPERATING
PROCESSOR(S) I
STORAGE
1420
: ....................................
OSYSTEM J;..
w / /_
1408_|
:
7|\
PROGRAM
: |NPUTDEV|CE(S)
1422

I | MODULES
1410~I|\1
I

PROGRAM

DATA

|
': OUTPUT DEV|CE(S) //
I.

1424

:,x /_

| COMMUNICATION /

\_ _______________________ _,'

CONNECT|ON(S)

1426
OTHER COMPUTING
DEVICES

Apr. 18, 2013

US 2013/0097697 A1

SECURITY PRIMITIVES EMPLOYING HARD

ARTIFICIAL INTELLIGENCE PROBLEMS

image in order to remember its location. A large set including

BACKGROUND

from the click-points used by people to build a dictionary,

Which, albeit much larger than that for other click-based
graphical passWord schemes, can then be used to launch
dictionary attacks to ?nd Weak passWords in PCCP.

[0001] It is a fundamental method in computer security to

create cryptographic primitives based on hard mathematical

problems that are computationally intractable. For example,

the problem of integer factorization is fundamental to RSA
(Which stands for Rivest, Shamir and Adleman Who ?rst

publicly described it) public-key cryptosystem and Rabin

encryption. The discrete logarithm problem is fundamental to
ElGamal encryption, Dif?e-Hellman key exchange, Digital
Signature Algorithm, elliptic curve cryptography and so on.

[0002]

Using hard arti?cial intelligence problems for secu

rity is an exciting neW paradigm. Under this neW paradigm,

the most notable primitive invented is CAPTCHA, Which is
noW a standard Internet security technique Where CAPTCHA
stands for Completely Automated Public Turing test to tell

Computers and Humans Apart. CAPTCHA distinguishes

betWeen humans and computers by presenting a challenge,
i.e., a puZZle, beyond the capability of computers but easy for
humans. Visual CAPTCHAs can be classi?ed into text

CAPTCHAs and image recognition CAPTCHAs depending

on What to recogniZed, characters or visual objects and

images. CAPTCHA has been largely deployed to protect

online email services from being abused by bots. Another
notable application is in a class of security protocols, herein
referred to as CAPTCHA-based PassWord Authentication

(CPA) protocols, in Which CAPTCHA and passWord are

combined in a protocol to defend against online dictionary
attacks on passWords.

[0003]

Another notable invention is POSH (PuZZle Only

Solvable by Human), Which is similar to CAPTCHA but the

ansWers are required to be consistent only for the same per
son. Different people may provide different ansWers to the

same puZZle. POSH Was proposed to mitigate o?iine dictio

nary attack on passWords or passWord-derived keys.
[0004] Pixels in an image are not equal. Some spots are
salient and informational, and thus more helpful in memoriZ

ing their locations than others. Hotspots and security of click

based graphical passWords have been studied. In general,
selecting hotspots in a passWord helps memoriZation of the
passWord but makes the passWord vulnerable to dictionary
attacks Wherein hotspots are collected to guess passWords. A
trial and error procedure is then applied to ?nd out the actual

passWords. Selecting more randomly distributed click-points

produces a stronger passWord but may lead to dif?culty to
remember it. This con?ict betWeen passWord memoriZation
and security presents a fundamental problem in designing a

graphical passWord system. Existing designs of click-based

all these salient structural points, i.e., extended hotspots,

can be collected by using image processing techniques and

SUMMARY

[0006] This disclosure describes a neW security primitive

based on hard arti?cial intelligence problems, namely, a fam

ily of graphical passWord systems. In some embodiments, the

graphical passWord systems may be built on top of
CAPTCHA technology, and such a graphical passWord sys
tem may be referred to as CaRP (CAPTCHA as gRaphical

PassWords) system.
[0007] A CaRP system employs a user selection-based
graphical passWord scheme, Where a user provides a
sequence of selections, such as by clicking on and/or touching
an image, and these selections are used to derive the users

passWord. HoWever, images used in a CaRP system may be

automatically generated by a CAPTCHA-based generator.

That is, the image upon Which a user makes selections in
CaRP may be a CAPTCHA challenge, and in some embodi
ments, a neW CaRP image is generated and/ or provided to the
user for every login attempt.

[0008] CaRP employs an approach very different from

existing click-based graphical systems. Goals of CaRP
approach include: 1) de-correlation of hotspots betWeen dif
ferent login attempts; and 2) disabling computers from cor
rectly inputting a graphical passWord or recovering the pass
Word from a users selections. The CaRP approach alloWs

using hotspots in a passWord for easy memoriZation but, in

some embodiments, varies the image for each login attempt,
and furthermore, the hotspots in the image of one login
attempt are made computationally uncorrelated to the

hotspots in the image of another login attempt to thWart

dictionary attacks. HoWever, the CaRP approach retains some
invariants that humans can use as a passWord but cannot be

computed by a computer. This constraint is similar to that of

CAPTCHA. By generating images satisfying this constraint,

these images can also be used to distinguish betWeen humans

and computers.
[0009] This Summary is provided to introduce a selection
of concepts in a simpli?ed form that are further described

beloW in the Detailed Description. This Summary is not

intended to identify key features or essential features of the
claimed subject matter, nor is it intended to be used to limit
the scope of the claimed subject matter.

graphical systems have tried to strike a balance betWeen the

tWo con?ict requirements.

[0005] Cued Click Points (CCP) enlarges the search space
by requiring a user to click a sequence of images in entering
a passWord instead of a single image commonly used graphi
cal passWord systems. Persuasive Cued Click Points (PCCP)
extends CCP by requiring a user to select a point inside a

randomly positioned vieWport in creating a passWord, result

ing in more randomly distributed click-points. These schemes
have mitigated the hotspot problem but still focused on strik
ing a balance betWeen passWord memoriZation and security.
They may still be vulnerable to dictionary attacks. For

example, the click-point selected from a randomly positioned

vieWport in PCCP may still be a salient structural point in the

BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The detailed description is described With reference

to the accompanying ?gures. In the ?gures, the left-most
digit(s) of a reference number identi?es the ?gure in Which
the reference number ?rst appears. The same reference num

bers in different ?gures indicate similar or identical items.

[0011] FIG. 1 is a schematic diagram of an illustrative
environment for a CaRP system.
[0012] FIG. 2 is a schematic diagram of a second illustra
tive environment for a CaRP system.
[0013] FIG. 3 is a schematic diagram of a third illustrative
environment for a CaRP system.

Apr. 18, 2013

US 2013/0097697 A1

[0014] FIG. 4 is a block diagram of illustrative relation

ships in a CaRP environment.
[0015] FIGS. 5A and 5B are block diagrams of illustrative
passwords in a CaRP environment.
[0016] FIG. 6 is a block diagram of an illustrative user

interface displaying a primitive domain.

[0017] FIG. 7 is How diagram of an illustrative process for
providing a user With a passWord in a CaRP system.

[0018]

FIG. 8 is How diagram of an illustrative process for

implementing CaRP With intemet protocols.

[0019]

FIG. 9 is a block diagram of another illustrative

display device displaying a CaRP image.

[0020] FIG. 10 is a block diagram of a primitive domain
that may be employed in a CaRP environment.
[0021] FIG. 11 is a How diagram of an illustrative proce

dure for generating a CaRP ClassObjs Image.

[0022]

FIGS. 12A-12C are block diagrams of an illustrative

display displaying a CaRP ClassObjs image.

[0023]

FIGS. 13A and 13B are diagrams of illustrative

point-based ?ne point CaRP (PP-CaRP) potential passWord

elements.
[0024]

FIG. 14 is a block diagram of an illustrative com

puting device that may be deployed in the environment shoWn

in FIGS. 1-3.

relayed to humans, the humans ansWer the CAPTCHA

challenges, and the humans ansWers are returned.
[0029] CaRP can also mitigate spyWare threats and can be

applied to dramatically reduce spamming emails sent from an

online email service provider.
[0030] CAPTCHA-based PassWord Authentication (CPA)
protocols bind a CAPTCHA and a passWord together in an
authentication process. In these protocols, the CAPTCHA
and the passWord are still tWo separate entities. HoWever, in
some embodiments, the intrinsic combination of the
CAPTCHA and the passWord in CaRP creates a neW security
primitive. In some embodiments, CaRP may be both a

CAPTCHA and a passWord authentication system, offering

not just defense against online dictionary attack, but also

additional advantages such as being robust to relay attacks on
CAPTCHAs that CPA protocols, using CAPTCHAs to
defend against online dictionary attacks, are vulnerable to.
[0031] In some embodiments, CaRP may be both a
CAPTCHA and a graphical passWord system. In some
embodiments, CaRP may use a domain of potential pass

Word-elements, each potential passWord-element being an

image of an object (e.g., characters, animals, manmade
objects, mythical creatures, etc.) to generate, at each login
attempt, a CAPTCHA image based on some or all of the

potential passWord-elements. Each potential passWord-ele

DETAILED DESCRIPTION

ment in the domain may have different incarnations. For

OvervieW

may be appear as Times NeW Roman, Arial, or other fonts; or

an animal such as a dog may appear to be different breeds
such as a poodle, a Saint Bernard, Australian Shepherd, Bor

[0025] Conceptually, a CaRP (CAPTCHA as gRaphical

Passwords) maybe seem uncomplicated but it may be
generic, and it may have multiple instantiationsia CaRP
system can be built on top of either a text CAPTCHA or an

image recognition CAPTCHA. For example, one of such

example, a potential passWord-element of a character A

der Collie, etc. Object recognition is needed in CaRP. As With

CAPTCHA, the difference betWeen a human s capability and

a computers capability to recogniZe the objects in a CaRP

image makes object recognition an easy task for humans but

instantiations is a text CaRP, Where each image is a text

intractable for computers. CaRP thWarts dictionary attacks by

CAPTCHA challenge With printable symbols displayed in a

preventing building a useful dictionary to attack and auto

distorted form With sophisticated connections. In this

instance, the user remembers the users passWord as a text

mated attacks by preventing computers from knoWing hoW to

correctly input a passWord or deducing the actual passWord

passWord. HoWever, the user enters the passWord in a CaRP

even When the users selections are knoWn.

system by selecting the right character sequence on the

image. The user may select characters by, among other things,

[0032] Among other things, a signi?cant difference

betWeen CaRP and existing selection/click-based graphical
passWord systems is that While hotspots still exist in each
CaRP images, hotspots cannot be exploited to launch dictio
nary attacks against CaRP. Hotspots in different CaRP images

clicking on characters With a mouse and or touching a touch

sensitive input device.

[0026] CaRP thWarts automated attacks by preventing
computers from correctly inputting a pas sWord or recovering
the passWord from a users selections. Furthermore, CaRP

are computationally uncorrelated. This de-correlation makes

images used in different login attempts are computationally

a dictionary built upon collected hotspots from previously

used images useless in ?nding the hotspots in a neW image

uncorrelated, Which therefore effectively removes vulner

used in a current login. Dictionary attacks are therefore

ability caused by image hotspots, a problem inherent in many

selection/click-based graphical passWord schemes that often

memorize a passWord in CaRP, as discussed beloW. CaRP has

leads to Weak passWord choices. In particular, CaRP offers

protection against the folloWing attacks:

[0027] 1. Online dictionary attacks on passWords: These
have been a longstanding and major security threat for
various online services. This threat is noW Widespread
and considered as a top cyber security risk as more and
more online services use Web-based passWord authenti

thWarted. On the other hand, hotspots can be used to help

thus fundamentally removed the con?ict betWeen passWord

memoriZation and security that the existing graphical pass
Word systems try to strike a balance.

[0033] A passWord in CaRP is a sequence of visual object

names or primitive elements, e. g., images of characters (same
as conventional text passWord) for text-based CaRP. A dic
tionary attack on the sequence of visual object names or

cation.
[0028] 2. Relay attacks on CAPTCHAs: These have
been an increasing threat to online applications
defended by CAPTCHAs. For example, a computer
Worm, Koobface, Was used to break CAPTCHA security

primitive elements cannot be successful since, given a trial

sequence, a computer cannot correctly select/click the objects
in the sequence on the CaRP image to input a passWord due to

at FACEBOOK in order to register neW FACEBOOK

accounts. In a relay attack, CAPTCHA challenges are

recogniZe the visual objects in each CaRP image for each

login attempt. A human-based dictionary attack is signi?

its inability to recogniZe the visual objects in the imageian

attribute of CAPTCHA. Humans have to be employed to

Apr. 18, 2013

US 2013/0097697 A1

cantly inef?cient and usually costly since humans are much

slower to complete one login attempt than are computers.
[0034] There are tWo major differences betWeen a CaRP

protected to ensure the security of the communicated mes

sages. In another embodiment, the user device 104a may
locate the points in a set a priori that are closest to and Within

image and a CAPTCHA challenge image:

[0035] 1. All the visual objects in the primitive domain

a tolerable error of the selected coordinates [(x1,y1), (x2, y2)

may, or should, appear in a CaRP image but are not

information based at least partially on these located points.

. . . (xK, yK)], and sends to the security administrator 102

necessarily in a CAPTCHA challenge image. This

Based at least on the information received from the user

alloWs a user to enter any possible passWord.

device 10411, the security administrator 102 determines

Whether the user 12411 selected regions or points correspond
ing to the users passWord.

[0036] 2. Visual objects in a CaRP image can be arranged

randomly Without any order. Visual objects in a
CAPTCHA image, on the other hand, may be of certain
order so that a user knoWs hoW to type in a proper order.

For example, for a text-based CaRP image characters

may be arranged anyWhere on a 2D space but only from
left to right for a text CAPTCHA challenge.

[0041] The user device 1041) may be a portable user device

such as, but not limited to, a cell phone, a smart phone, a

tablet, a personal digital assistant, and may be a multifunction

device. The user device 1041) includes user-interface compo
nent 126. The user interface component 126 includes a dis

Illustrative Environments

play component 128, Which may be a touch sensitive compo

nent, and may include keys 130.

[0037] FIG. 1 is a schematic diagram of an illustrative

environment 100 for employing CaRP. The environment 100

security administrator 102. The user device 1041) is in Wire

[0042]

A user 1241) uses the user device 1041) to access the

may include at least one security administrator 102, and one

less communication, via Wireless communication link 132,

or more user devices, individually referenced as 10411 and

With a netWork 134, and a communications link 136 commu

10419 and collectively referenced as 104.

nicatively couples the netWork 134 to the security adminis

[0038] The security administrator 102 may be a computing

system such as a server, Web server, desktop computer, laptop

trator 102.

computer, or the like. The security administrator 102 is com

provides an image to the user device 1041) for display on the

display component 128. The user 1041) uses the user interface
component 126 to select one or more regions or points of the

municatively coupled to security primitive database 106 and

to user pro?le database 108, via communication links 110.

The security primitive database 106 may include, among

other things, images that are used to generate log-in chal
lenges that are displayed to users of the CaRP. The user pro?le
database 108 may include, among other things, user identi?
cations (user-ids) of users of the CaRP. The user pro?le data
base 108 may further include a respective passWord indicator

[0043] Among other things, the security administrator 102

image displayed on the display component 128. The user

1241) may select a region of the images by, among other
things, using the keys 130 to position a cursor over a point
inside the region to select and to click upon the point.
Similarly, the user 1241) may select a region of the image by,
among other things, touching the display component 128 on

example of a passWord indicator is hash value H(P,s) of a

a point Within the region to select. The user 1241) may select
a point of the image in a similar manner. The user device 1041)
collects the coordinates [(x1,y1), (x2, y2) . . . (xK, yK)] of the

users passWord P With a salt s. The user pro?le database 108

may further include a respective passWord for users of the

points that the user input to the interface component 126, and
sends to the security administrator 102 information based at

for users of the CaRP in Which a passWord indicator may

indicate a passWord but is not a passWord. A non-limiting

CaRP. The passWords may be encrypted for security pur

least on the points the user input. In one embodiment, the user

poses.
[0039] The user device 104a may be a desktop computing
system having a toWer 112 and user interface devices 114.

device 1041) sends the coordinates [(x1,y1), (x2, y2) . . . (xK,

The user interface devices 114 may include a mouse and/ or a

yK)] to the security administrator 102. This communication

may be protected to ensure the security of the communicated
messages. In another embodiment, the user device 1041) may

touchpad 116, a keyboard 118 and display device 120.

locate the points in a set a priori that are closest to and Within

[0040]

a tolerable error of the selected coordinates [(x1,y1), (x2, y2)

A user 12411 uses the user device 10411 to access the

security administrator 102 via a Wired or Wireless communi

. . . (xK, yK)], and sends to the security administrator 102

cation link 122. Among other things, the security administra

information based at least partially on these located points.

tor 102 provides an image to the user device 10411 for display
on the display device 120. The user 12411 uses the user inter
face devices 114 to select one or more regions or points of the

image displayed on the display device 120. The user 12411

may select a region of the images by, among other things,

using the mouse and/ or touchpad 116 and/ or keyboard 118 to
position a cursor over a point inside the region to select and to
click upon the point. The user 12411 may select a point by,

Based at least on the information received from the user

device 104b, and the security administrator 102 determines

Whether the user 1241) selected regions or points correspond
ing to the users passWord.
[0044] FIG. 2 is a schematic diagram of an illustrative
environment 200 for employing CaRP. The environment 200
includes a computing system 202 such as a personal com

puter, desktop computer, laptop computer, server, Web server,

among other things, using the mouse and/or touchpad 116

etc. The computing system 202 may include a toWer 204 and

and/or keyboard 118 to position a cursor over the point to

select and to click upon the point. The user device 104a
collects the coordinates [(x1,y1), (x2, y2) . . . (xK, yK)] of the

user interface devices 206. The user interface devices 206

may include a mouse and/ or a touchpad 208, a keyboard 210

104a sends the coordinates [(x1,y1), (x2, y2) . . . (xK, yK)] to

and display device 212.

[0045] The computing system 202 includes a security mod
ule 214. The security module 214 includes security primitives
216 and user pro?les 218. The security primitives 216 may
include, among other things, images that are used to generate

the security administrator 102. This communication may be

log-in challenges that are displayed to users of the CaRP. The

points that the user input to the interface devices 114, sends to
the security administrator 102 information based at least on
the points the user input. In one embodiment, the user device

Apr. 18, 2013

US 2013/0097697 A1

user pro?les 218 may include, among other things, user iden
ti?cations (user-ids) of users of the CaRP. The user pro?les
218 may further include a respective password indicator for
users of the CaRP in Which a passWord indicator may indicate
a passWord but is not a passWord. A non-limiting example of
a passWord indicator is hash value H(P,s) of a user s passWord
P With a salt s. The user pro?les 218 may further include a

respective passWord for users of the CaRP. The passWords

may be encrypted for security purposes.
[0046]

A user 220 access the computing system 202 via the

users passWord based at least on the coordinates [(x1,y1),

(x2, y2) . . . (xK, yK)] of the points that the user selected.

[0052]

FIG. 4 is a block diagram of relationships in a CaRP

environment 400. Referring to FIGS. 1-3, the security admin

istrator 102 or the security module 214/310 generates a CaRP
image and provides the CaRP image to a user-device such as
user-devices 104a, 104b, 302 or computing system 202. The
CaRP image is based on images in a primitive domain 402.

The primitive domain 402 includes N potential passWord

elements (PPEs) (PPE1-PPE N) 404. Collectively, the poten

security module 214. Among other things, the security mod

tial passWord-elements 404 span all possible passWords in the

ule 214 provides an image for display on the display device

CaRP environment 400.

212. The user 220 uses the user interface devices 206 to select

[0053] A displayed image 406 is provided to the user via the

user-device. The displayed image 406 includes a number of

one or more regions or points of the image displayed on the

display device 212. The user 220 may select a region of the

passWord-element indicators (PEI 1-PEI M) 408. Each pass

images by, among other things, using the mouse and/ or touch

Word-element indicator 408 corresponds to one of the poten

pad 208 and/or keyboard 210 to position a cursor over a point

tial passWord-elements 404. For example, PEI 1 corresponds

inside the region to select and to click upon the point.

Similarly, the user 220 may select a region of the images by,
among other things, using the mouse and/or touchpad 208

to PPE 1, PEI 2 corresponds to PPE 2, and so on. In some

select and to click upon the point.

[0047] The security module 214 determines Whether the
user 220 selected regions or points corresponding to the
users passWord based at least on the coordinates [(x1,y1),

embodiments, the number M of passWord-element indicators

408 may be equal to the number N of potential passWord
elements 404. HoWever, in other embodiments, the number M
of passWord-element indicators 408 may be less/ greater than
the number N of potential passWord-elements 404. For
example, in one embodiment, tWo passWord-element indica
tors in the displayed image 406 may be of the same potential

(x2, y2) . . . (xK, yK)] of the points that the user selected.

pas sWord-elements but are visually different in order to make

[0048] FIG. 3 is a schematic diagram of an illustrative

environment 300 for employing CaRP. The environment 300
includes a portable user device 302 such as, but not limited to,
a cell phone, a smart phone, a tablet, a personal digital assis
tant, and a multifunction device. The user device 3 02 includes
user-interface component 304. The user interface component
304 includes a display component 306, Which may be a touch

it more dif?cult for computers to recogniZe the passWord

elements in the displayed image 406. This may lead to M>N.
In another embodiment, an a priori subset of the potential
passWord-elements is used in a passWord, and the passWord
element indicators corresponding to the subset of PPEs
appear in the displayed image 406. This may lead to M<N.

and/or keyboard 210 to position a cursor over the point to

sensitive component, and may include keys/buttons 308.

[0049]

The user device 302 includes a security module 310.

The security module 310 includes security primitives 312 and

user pro?les 314. The security primitives 312 may include,
among other things, images that are used to generate log-in
challenges that are displayed to users of the CaRP. The user

Typically, each passWord-element indicator 408 is displayed

to be visually different from its corresponding potential pass
Word-element 404, and tWo passWord-element indicators 408
corresponding to the same potential passWord-element 404
and displayed in the same displayed image 406 or tWo differ

ent displayed images 406 are displayed to be visually differ

ent. The visual difference may be due to different incarnations

pro?les 314 may include, among other things, user identi?

of the potential passWord-element 404 and/or through opera

cations (user-ids) of users of the CaRP. The user pro?les 314

may further include a respective passWord indicator for users
of the CaRP in Which a passWord indicator may indicate a
passWord but is not a passWord. A non-limiting example of a

tions, joint or independent, such as, but not limited to, uni

form and/or non-uniform scaling, rotation, changing of point

of-vieW, deformations, transformations and/or CAPTCHA

operations.

passWord indicator is hash value H(P, s) of a users passWord

[0054]

P With a salt s. The user pro?les 314 may further include a

number (K) passWord-elements (PE) 412. For the sake of

clarity, passWord 410 is shoWn as consisting of four passWord

respective passWord for users of the CaRP. The passWords

may be encrypted for security purposes.
[0050]

A user 316 accesses the user device 302 via the

security module 310. Among other things, the security mod

ule 310 provides an image (not shoWn) for display on the
display component 306. The user 316 uses the user interface
component 304 to select one or more regions or points of the

image displayed on the display component 306. The user 316

may select a region of the images by, among other things,
using the keys/buttons 308 to position a cursor over a point

inside the region to select and to click upon the point. The
user 316 may select a point by, among other things, using the
keys/buttons 308 to position a cursor over the point to select
and to click upon the point. Similarly, the user 316 may

select regions or points of the image by, among other things,

touching the display component 306.
[0051]

A users passWord 410 may be a sequence of a

elements PE 1-PE 4. HoWever, other passWords may include

feWer, the same, or a greater number of passWord elements
412. Each one of the passWord-elements 412 corresponds to

one of the potential passWord-elements 404. For example, PE

1 corresponds to PPE1, PE 2 corresponds to PPE 4, PE 3
corresponds to PPE 5 and PE 4 corresponds to PPE N.
[0055] The user selects points/regions, individually refer
enced as 414a-414d and collectively as 414, ofthe displayed
image 406. User input 416 corresponding to the users
selected points/regions 414 is provided to or used to produce
information to send to the security administrator 102 or the
security module 214/310. The user input can be sent directly
to security administrator 102 or the security module 214/310
(When a communication channel betWeen the user-devices

104a, 104b, 302 or computing system 202, and the security

The security module 310 determines Whether the

administrator 102 or the security module 214/310 is secure)

user 316 selected regions or points corresponding to the

or is used to deduce information (such as in a challenge

Apr. 18, 2013

US 2013/0097697 A1

response authentication Which is typically used When the

communication channel is not secure) to send to the security
administrator 102 or the security module 214/310. A chal
lenge-response authentication protocol may be used to
exchange information betWeen tWo parties such that one party
tries to convince the other party that he/ she knoWs the pass

Word Without actually sending the passWord.

determines Whether the purported passWord-elements corre

spond, in type and in sequence, to the actual the passWord
elements 412.

[0059]

In another embodiment, security administrator 102

or the security module 214/310 derives the sequence of

selected points/regions corresponding to the passWord 410 of

the user stored in the user pro?le database 108 or user pro?les

The user input 416 is a sequence of K user selections

218/314, and compares With the received information to

determine if the user input 416 matches the sequence derived

(US) 418. For the sake of clarity, user input 416 is shoWn as
consisting of four user selections US l-US 4. However, other
user inputs may include feWer, the same, or a greater number

from the passWord 410.

[0060] FIGS. 5A and 5B are block diagrams of passWords
502a and 502b, respectively. Each one of the passWords 502a

[0056]

of user selections 418. Each user selection 418 provides a

and 50219 is a sequence of a number of passWord-elements.

coordinate Q(,Y) of a point corresponding to the user s selec

tion. In some embodiments, a passWord-element indicator
408 may have a point of coordinate (X,Y) associated With it,
and When the user selects the pas sWord-element indicator 408
the coordinate Q(,Y) is provided as one of the user inputs 416.
In some embodiments, there may be a set of selectable points/

PassWord 50211 is comprised of passWord-elements 504-1

through 504-N, Which are images of alphanumeric charac
ters. An image having the passWord 502a therein Will include
images of the characters A, 7, and Q, and all other
passWord-elements betWeen 504-1 and 504-N.
[0061] Password 502!) is comprised of passWord-elements

regions associated With the passWord-element indicators 408

displayed in the displayed image 406, and each selectable
point/region 414 may have a coordinate, e.g., Q(a,Ya), asso
ciated With it, and When the user selects a point of coordinate

(X,Y), the closest selectable point/region is located, and its

coordinate, e.g., Q(a,Ya), is provided as one of the user selec
tions 418. In some embodiments, the distance betWeen the

506-1 through 506-N, Which are images of objects such as a

car, tree and building. An image having the passWord 502b

therein Will include images of the car, tree, and build
ing, and all other passWord-elements betWeen 506-1 and
506-N. As another example, a passWord may be comprised of
a number of passWord-elements that are images of objects
such as a dog, a cat, a tiger and/or other animals Which are
similar to one another. Images that are similar to one another

user-selected point of coordinate Qi, Y), and the coordinate

(Xa, Ya) of the closest selectable point/region should be

may be used because humans may easily discern the differ

Within a tolerable error. In some embodiments, When the user

differentiate one from another.

selects a point/region 414 of the passWord-element indicator

408, the coordinate Q(,Y) of the selected point is provided as

ences betWeen the images but computers have a hard time to

[0062]

In some embodiments, a passWord may be com

one of the user selections 418.

prised of multiple types of passWord-elements that may be a

combination of images of different types of objects, for

[0057]

example, animals and alphanumeric characters.

The security administrator 102 or the security mod

ule 214/310 receives information based at least on the user

input 41 6. In one embodiment, the received information is the

user input 416. This is typically applied When the security

administrator 102 or the security module 214/310 can receive
the user input 416 in a secure manner such as via secure

communication channel. In another embodiment, the

received information may be different from the user input
416, but is based at least on the user input 416. Even When the
information sent to the security administrator 102 or the secu

rity module 214/310 is knoWn to adversaries, the user input

still cannot be deduced. This is typically used When the com
munication channel to the security administrator 102 or the
security module 214/310 is not secure. The user input 416 is
a purported passWord, Which may or may not be valid. Based

at least on knowledge of the displayed image 406 and the

received information, the security administrator 102 or the
security module 214/310 determines if the user input 416
matches the passWord 410 or its passWord indicator (not
shoWn) of the user, Which may be stored in the user pro?le
database 108 or user pro?les 218/314.

[0058] In one embodiment, the security administrator 102

or the security module 214/310 receives the user input 416,
and determines Which passWord-element indicators 408 cor
respond to the sequence of user selections 418, and maps
these passWord-element indicators 408 back to the corre

sponding potential passWord-elements 404. These corre

sponding potential passWord-elements 404 are considered

Creating a PassWord
[0063] FIG. 6 is a block diagram of a user-interface 600
displaying a primitive domain 602. The user-interface 600
may be vieWed on the user-devices 104a, 104b, 302 (see

FIGS. 1 and 3) and the computing system 202 (see FIG. 2).
[0064] The primitive domain 602 includes multiple poten
tial passWord-elements 604. The multiple potential pass
Word-elements 604 span the space of all possible passWords.
Each one of the potential passWord-elements 604 is an image
that can be selected as a passWord-element. In the illustrated

embodiment of the primitive domain 602, the potential pass

Word-elements 604 are alphanumeric characters. Some com

mon alphanumeric characters such as letters I (capital i)

and O are omitted due to their visual similarities to numbers
1 and 0.

[0065]

FIG. 7 is a How diagram ofa process 700 for pro

viding a user With a passWord in a CaRP system. The process

700 is illustrated as a collection of blocks in a logical ?oW

graph, Which represent a sequence of operations that can be

implemented in hardWare, softWare, or a combination
thereof. In the context of softWare, the blocks represent com
puter-executable instructions that, When executed by one or
more processors, cause the one or more processors to perform

the recited operations. Generally, computer-executable

purported passWord-elements, Which may or may not corre

instructions include routines, programs, objects, compo

nents, data structures, and the like that perform particular
functions or implement particular abstract data types. The

spond to the passWord-elements 412 of the passWord 410. The

security administrator 102 or the security module 214/310

order in Which the operations are described is not intended to

be construed as a limitation, and any number of the described

Apr. 18, 2013

US 2013/0097697 A1

blocks can be combined in any order and/or in parallel to

[0076]

implement the process. Other processes described throughout

this disclosure shall be interpreted accordingly.
[0066] At 702, a primitive domain of potential passWord

CaRP for Web Application, may be implemented to Work

With the Widely used HTTP Basic Authentication. Note that
for both HTTP authentication protocols, TLS/SSL provides

The CaRP schemes in some embodiments, such as

elements is de?ned. Collectively, the potential passWord-el

the communication security betWeen a client and a Web

ements span all possible passWords. The CaRP environment,

server.

such as the security primitive database 106 in FIG. 1 and

security primitives 216 and 312 in FIGS. 2 and 3, may contain

[0077] FIG. 8 is a How diagram of an illustrative process

800 for implementing CaRP With intemet protocols such as
HTTP Basic Access Authentication. The process 800

multiple incarnations for each potential passWord-element.

For example, the CaRP environment may contain Times NeW
Roman, Arial, or other fonts for a potential passWord-element
of a character A; or a Saint Bernard, Australian Shepherd,
Border Collie, etc. for a potential passWord-element of an

animal dog. Different incarnations of a potential passWord

element introduce more variations in the generated CaRP

images, making it harder for computers to recogniZe.

[0067] At 704, a user registers With the CaRP system. For
example, referring to FIG. 1, user 12411 may register With the

security administrator 102. Similarly, referring to FIGS. 2/3,

user 220/316 may register With the security module 214/310.

Registration may include providing identifying information

and creating/establishing/providing a user-id.

[0068] At 706, the user is provided With potential pass

Word-elements. The potential pas sWord-elements may be dis

played on user-devices 104a, 104b, 302 (see FIGS. 1 and 3)

and computing system 202 (see FIG. 2). In some embodi
ments, an entire primitive domain may be displayed at one
time. In other embodiments, one or more of the potential

pas sWord-elements of a primitive domain may be displayed at

one time.

[0069] At 708, user input for selected potential passWord

elements is received. In some embodiments, the user may

select, one-by-one, a number of potential passWord-elements,

in Which each selected potential passWord-element is then
assigned to a corresponding one of the passWord-elements.
[0070] At 710, Which in some embodiments may be
optional, a passWord indicator may be generated. The pass
Word indicator may be indicative of the corresponding pass
Word and may be used for verifying that a purported passWord
entered by a user is in fact a valid passWord. For example, a
passWord indicator may be a hash value H(P,s) of a users
passWord P With a salt s.

[0071] At 712, the users passWord and/or the generated

passWord indicator is associated With the user. The passWord
and/ or the generated passWord indicator may be stored in the

security primitive database 106 and/ or security primitives

216/ 312.

[0072]
[0073]

CaRP for Web Applications

Security of user authentication in Web applications

involves acts implemented at a client 802 and a security

administrator 804, Which may be embodied in a Web server.

The security administrator 804 may store user-IDs and pass

Word indicators.
[0078] At 806, the security administrator 804 receives an
access request from the client 802. The access request may,

optionally, include the users user-id.

[0079] At 808, the security administrator 804 generates a
CaRP image. In some embodiments, the CaRP image
includes all of the potential passWord-elements that make up
the primitive domain 602. In other embodiments, the CaRP
image includes less than all of the potential passWord-ele
ments that make up the primitive domain 402, 602.
[0080] At 810, the security administrator 804 sends the
CaRP image to the client 802.A user selects regions/points on

the received image to enter the users passWord. For example,

the user may select characters A, B, C, D, #, 9,
8, and 7 sequentially on the CaRP image.
[0081]

At 812, the coordinates [(x1,yl), (x2, y2) . . . (xK,

yK)] of the selected points on the CaRP image are provided to

the security administrator 804. In some embodiments, the
users user-ID is also provided, using the HTTP Basic
Authentication, to the security administrator 804 With the
coordinates [(xl,yl), (x2, y2) . . . (xK, yK)] of the selected

points on the CaRP image.

[0082] At 814, the security administrator 804 determines
Whether to grant or deny access. The security administrator
804 may map the coordinates back on the CaRP image pre
viously provided to the client 802, and recover a sequence of
purported passWord-elements, P', that the user selected on the
CaRP image at the client 802. The security administrator 804
may then determine Whether the sequence of purported pass

Word-elements, P', correspond to the sequence of passWord

elements that make up the users passWord. For example, the
security administrator 804 may retrieve the salt s of the asso
ciated With the user-id, calculate a hash value of P' With the
salt, and compare the result With a hash value stored for the
account.

[0083]

At 816, access is granted only ifthe purported pass

Word is veri?ed to be the same as the actual passWord, e.g.,

relies on tWo parts: Hypertext Transfer Protocol Secure (HT

TPS) and HTTP Authentication.

hash value of the purported passWord agrees With the hash

value stored for the account. If the security administrator 804
denies access, then the process may return to 808, Where

[0074] HTTPS. The Transport Layer Security (TLS) or its

predecessor, Secure Sockets Layer (SSL), provides server
authentication, con?dentiality, and integrity at the transport

another, and different, CaRP image may be generated. Each

generated CaRP image is computationally uncorrelated With
another CaRP image.

layer. It builds a secure communication channel betWeen a

client and a Web server.

[0084] The security administrator 804 may deny access

after a number of failed attempts. For example, after three

[0075] HTTP Authentication. HTTP Authentication runs at

the application layer to authenticate a client to a Web server.

administrator 804 may determine, at 814, to deny access.

failed attempts, instead of repeating 808-812, the security

This is one protocol Wherein CaRP may be used. Web appli

cations almost exclusively adopt the HTTP Basic Access
Authentication protocol for user authentication, Wherein the
client sends both userID and passWord, encoded With Base64

administrator 804 is not required to store the user passWords.

to the server for user authentication.

Instead, only a passWord indicator for a passWord may be

[0085]

In some embodiments, the security administrator

804 has the capability to recover a users passWord P in a

successful authentication procedure. HoWever, the security

Apr. 18, 2013

US 2013/0097697 A1

recorded. Further security measures can be taken to make it

passWord-element indicator. The resulting bounding boxes

hard or impossible for the security administrator 804 to

recover passWords. For example, threshold cryptography can
be applied so that the security administrator 804 knows only
a part of a users passWord and/or passWord indicator and
other computing devices knoW the remainder.

are used by the security administrator/module to identify the

selected passWord-element indicators from the received
selection-points. Small overlapping may occur betWeen the

[0086]

However, in some embodiments, the security

bounding boxes of tWo neighboring passWord-element indi

cators. Preferably, users should not select a point very close to

the boundary of a passWord-element indicator to avoid click

ing on a fuZZy overlapping area. HoWever, this is generally

administrator 804 can store a passWord. The security admin

istrator 804 may store a passWord as a sequence of passWord

not an issue since overlapping areas make a tiny portion of a

elements, e.g., the passWord may be a sequence of images.

passWord-element indicators bounding box.

[0087] Alternatively, in some embodiments, each potential

passWord-element in a primitive domain may be assigned a

ClassObjs CaRP

passWord-element identi?er. The security administrator 804

may store a passWord as a sequence of passWord-element

identi?ers. Similarly, the security administrator 804 may

[0093] Image recognition CAPTCHAs can also be used to

build ClassObjs CaRP. The primitive domain used in Clas

store generate a passWord indicator based at least in part on

sObjs CaRP may be several classes of objects such as sets of

the sequence of passWord-element identi?ers.

Exemplary CaRPs
[0088] TextObjs and ClassObjs are tWo types of CaRP
Images or tWo CaRP schemes. TextObjs is an object-based

CaRP using a primitive domain consisting of alphanumeric

characters and printable symbols. ClassObjs is an object
based CaRP using a primitive domain consisting of objects in
different classi?cations. They may employ different
CAPTCHA schemes:

TextObjs may employ a text

CAPTCHA scheme While ClassObjs may employ an image

recognition CAPTCHA scheme.

different
For each type
typesof
ofanimal,
animals,
one
e.g.,
or multiple
{Dog, Cat,
3DPig,
models
Rabbit,
are built.
.. .

[0094] FIG. 10 shoWs a block diagram of a primitive

domain 1000 that may be employed in a ClassObjs CaRP
environment. The primitive domain 1000 includes a number
of classes 1002. Each class 1002 may be comprised of a
number ofsubclasses 1004 and subclasses 1004 may be com

prised of sub-subclasses 1006. A users passWord may be

comprised of multiple passWord-elements selected from dif

ferent classes, subclasses, or sub-subclasses. For example, as
described beloW, a users passWord may be comprised of a
number of passWord-elements selected from the class Life

Forms, subclass Mammals.

TextObjs CaRP

[0095]

[0089] In some embodiments, the primitive domain of Tex

tObjs CaRP may consist of alphanumeric characters and

generating a ClassObjs Image.

[0096] At 1102, the number of potential passWord-ele

printable symbols With confusing characters being removed

ments, from Which a ClassObjs image is generated, is deter

to avoid causing any recognition errors for humans. For

example, the letter O and numeral 0 (Zero) may appear
confusedly similar in a CaRP image, the numeral 1 (one)
and the letter I may also appear confusedly similar.
[0090] A passWord in this case is a sequence of passWord

elements in the primitive domain comprising alphanumeric

characters and printable symbols. Existing text CAPTCHA
can be modi?ed to generate TextObjs CaRP Wherein charac

ters (potential passWord-elements) in the alphabet (primitive

domain) are transformed into passWord-element indicators
and arranged randomly in tWo-dimension space in the CaRP

image. The passWord-element indicators correspond to

potential passWord-elements of the primitive domain, but the
passWord-elements indicators are presented in a different

visual representation than the potential passWord-elements

and are yet recogniZable to the user as the corresponding

potential passWord-element.
[0091]

FIG. 9 is an diagram of display device 900 display

ing a CaRP image 902. The CaRP image 902 has a number of
passWord-element indicators 904 arranged in a tWo-dimen
sional pattern. The passWord-element indicators 904 include
a number of valid passWord-element indicators 906-1
through 906-5 With the remainder of the passWord-element

FIG. 11 is a How diagram ofa procedure 1100 for

mined. For example, the ClassObjs image may be generated

based on four different potential passWord-elements such as

four types of mammals, e.g., Dog, Cat, Pig, and Rabbit. The
passWord-element indicators are randomly partitioned into
contributions, at least one, from each type of the potential
passWord-elements. For example, the number of passWord
element indicators may be ten, and the ten passWord-element
indicators may be portioned into sets of three Dogs, tWo Cats,
four Pigs, and one Rabbit. In this example, a user s passWord
is a sequence of mammals such as P:{Cat, Dog, Cat, Dog,

Pig, Dog}.
[0097] At 1104, 3D models of the potential passWord-ele
ments (e.g., Dog, Cat, Pig, Rabbit) are used to generate the
number of 2D passWord-element indicators for each potential

passWord-elements (e.g., 3 Dogs, 2 Cats, 4 Pigs, 1 Rabbit) by

applying different vieWs, textures, colors, lightning effects,
and optional distortion.
[0098] At 1106, the resulting passWord-element indicators
are then arranged. In some embodiments, the resulting pass
Word-element indicators are randomly arranged and/or in
some embodiments randomly arranged on a cluttered back

a users passWord is the sequence RMSAD. Then to cor

ground such as grassland. Some of the passWord-element

indicators may be occluded by other passWord-element indi
cators in the CaRP ClassObjs Image, but the major features

rectly enter the users passWord, the user must select, in the
proper sequence, valid passWord-element indicators 906-1

fying each entity in the image.

indicators being invalid passWord-element indictors. Assume

through 906-5.
[0092] During generation of the CaRP image 902, the secu
rity administrator/module may track a bounding box of each

should not be occluded to avoid confusing humans in identi

[0099]

At 1108, an image is generated in Which the pass

Word-element indicators are arranged, randomly or in another

manner, and Which may include a cluttered background.

Apr. 18, 2013

US 2013/0097697 A1

GridClassObjs CaRP
[0100]

FIGS. 12A-12C are block diagrams of an illustrative

display 1200 displaying GridClassObjs CaRP, Which com

bines ClassObjs CaRP With a content-sensitive grid.
[0101] FIG. 12A shows a ClassObjs image 1202 containing
a number of passWord-element indicators 1204. The Clas
sObjs image 1202 is shoWn in an unselected state. In this
state, a user has either not yet selected one of the passWord
element indicators 1204 or has ?nished the selection of one of

the passWord-element indicators 1204. The process of select

ing passWord-element indicators 1204 is described beloW.

[0102] When the CaRP ClassObjs image 1202 is received,
the user selects a passWord-element indicator 1204 that

matches the ?rst passWord-element of the users passWord.

For a passWord P:rabbit, . . . , suppose there are tWo

passWord-element indicators 1204 in image 1202 that are

rabbits, one in pink and the other in green, the user can select

a pink rabbit by selecting, via touching or clicking, a point

Word, the user selects a point inside the grid-square (1, 2), and
then selects a point inside the grid-square (3, 1) to ultimately
select the tWo grid-squares. The user then returns to the pass

Word-element indicator (animal image) to select the next the

passWord-element indicator (animal image) in the users

passWord. If there is no grid index folloWing the passWord
element indicator (animal image) in the user s passWord, such
as the second the passWord-element indicator Cat in the
example passWord, the user does not select any grid-square in
the n><n grid shoWn after a Cat is selected. The selected

points in selecting the passWord-element indicator (animal

image) and grid-squares form a sequence of coordinates, e. g.,

PPEI(150,50),GP(30, 66),GP(89, 160), PPEI(135, 97),

PPEI (82, 112), . . . , Where PPEI(x,y) means the coordi

nates of a point on the passWord-element indicator (animal

image), and GP(x,y) means the coordinates of a point on the
image of n><n grid. This sequence is sent to the security
administrator/module.

inside it. The user can equivalently select a green rabbit by

[0106] Upon receiving the sequence of point coordinates,

selecting, via touching or clicking, it.

[0103] Referring to FIG. 12B, a bounding rectangle 1206

the security administrator/module maps the ?rst PEPI on the

for the selected passWord-element indicator 1204 is calcu

lated using image processing techniques, for example Mean

Shift, and displayed. Due to occlusion and the cluttering
background, the calculation might be inaccurate. The user
may visually verify the correctness of the displayed rectangle
1206, and may drag an inaccurate side to roughly align With
a correct position if needed. Image processing techniques
may then be applied to re?ne the moved side to align With the

ClassObjs image 1202, and determines the selected pass

Word-element indicator, e.g., Rabbit. The security admin
istrator/module then uses the bounding box of the selected
passWord-element indicator to form a grid, and maps the

folloWing selected grid-points (GP) of grid image received on

the grid image to ?nd out the indices of the selected grid
squares, e.g., Grid(1,2),Grid(3, 1). Unlike the client side,
the server does not need to calculate the bounding rectangle of

nearby edges. This semi-automatic correcting procedure may

the selected animal. It records the bounding rectangle of each

animal in the image during generating the image. This pro

be repeated until the user is satis?ed With the accuracy of the

cess is repeated until all the select-points in the received

bounding rectangle. In most cases, the calculated bounding

sequence have been mapped, e. g., Rabbit, Grid(1,2), Grid(3,

rectangle is accurate and there is no need to correct anything.

1); Cat, Horse, Grid(2,2); Dog, . . . . This recoveredpassWord

[0104]

sequence is veri?ed such as by then performing a hash With

the salt of the account, and comparing the resultant hash value

Referring to FIG. 12C, once the bounding rectangle

1206 of the selected passWord-element indicator 1204 is

determined then a grid 1208 is displayed. The grid 1208 may
rely on the selected passWord-element indicator 1204. In

With a stored hash value.

some embodiments, the grid siZe is the same as the bounding

[0107] The number of grid-squares in a grid image is typi

cally selected to be larger than the siZe of the primitive

rectangle of the selected passWord-element indicator 1204.

domain. This leads to more choices on the grid-squares than

The grid 1208 may be approximately centered on the selected

on the primitive domain, resulting in a larger passWord space.

GridClassObjs may use a shorter passWord than ClassObjs
for the same level of security. This gain is at the cost of image
processing at the client side and much more complex selec
tions by users, leading to a longer time to select a passWord

passWord-element indicator 1204 and may appear superim

posed on the CaRP ClassObjs image 1202. The grid 1208
may also be placed at the center of the display. In the illus
trated embodiment, the grid 1206 is a 3x3 grid. In other

embodiments, the grid may be of different siZe (e. g., n><n) and
may be rectangular (e.g., m><n). A users passWord can be a

and less user-friendly.

sequence of passWord-elements 1204 interleaving With grid

TextPoints CaRP

indices, e. g., in a passWord Where passWord-element indica

tors are animals, the passWord may be P:Rabbit, Grid(1,2),
Grid(3, 1); Cat, Horse, Grid(2,2); Dog, . . . , Where Grid(1,2)

means the grid-square at index (1,2) and the grid indices after
an animal means the grid-squares With the grid determined by

the bounding rectangle of the preceding animal in the Clas

sObjs image 1202.

[0105] The passWord-element indicator 1204 inside the
selected bounding rectangle 1206 is displayed as a grid
square near the center of the grid 1208. In some embodiments,

the grid is displayed at the center of the display With the

passWord-element indicator 1204 displayed at the center of
the grid also moved to the center of the display. This ensures
that the screen has su?icient room to display all the grids for
the user to select. The user then selects a sequence of grid

squares that matches the grid indices folloWing the selected

passWord elements indicator 1204. For the example pass

[0108] It is also possible to design a point-based ?ne point

CaRP (PP-CaRP) scheme in Which multiple selectable points
inside an object can be used in a passWord. TextPoints is a
CaRP scheme that converts TextObjs to a FP-CaRP scheme.

It should be noted that ClassObjs can also be similarly

extended to a FP-CaRP scheme. In some embodiments,

salient structural points in an object may be used as selective

points. For example, TextPoints may use cross-points in a
character as selectable points. Salient structural points seem
to help in the folloWing Ways as compared to otherpoints in an

image:
[0109] 1. Long memory ofthe points in a FP-CaRP pass
Word With less recalling errors; and

[0110] 2. Higher accuracy in repeatedly selecting the

same point.

Apr. 18, 2013

US 2013/0097697 A1

[0111]

FIGS. 13A and 13B are diagrams of illustrative

potential password elements of TextPoints, a point-based ?ne

point CaRP (PP-CaRP) based on characters.
[0112] In one embodiment, only cross-points in a character
are alloWed as selectable points in a passWord. Referring to

FIG. 13A, an illustrative potential passWord element 1300 is

shoWn. A cross-point in a character is the point that tWo or
more strokes intersect. For example, the ?ve cross points (X)
1302 in FIG. 13A are cross-points of the letter A. These
cross-points are much easier to remember and to accurately

task can be accomplished, the underlying CAPTCHA is bro

ken: the characters in a CaRP image are recogniZed by com

puters since the image is actually a CAPTCHA challenge.

[0119] In generating a CaRP image, the server checks the
locations of the points in the primitive domain to make sure
that no point is occluded or too close to another character.
Upon receiving a CaRP image, a user identi?es the characters

in the image and selects her passWord points on the right

characters. A sequence of the coordinates of the selected
points is sent to the authentication server. Upon receiving the

locate than other points in A.

sequence, the server maps the coordinates on the image it sent

[0113]

to the client, and ?nds out the closest passWord candidate

points for each selected point and the error distance. If an

Different fonts can be used to generate a CaRP

image. For example, A of font Times NeW Roman can be

used in one CaRP image While A of font Arial can be used
in another CaRP image. A user has to select the same point
inside A no matter What alloWed font is used. This means
that in some embodiments the only alloWed points are invari
ant for different alloWed fonts. The ?ve cross points shoWn in

FIG. 13A are invariant points. In addition, points along the

error distance larger than a preset error tolerance is detected,

access is denied. Otherwise the sequence of passWord candi

date points is recovered from the received selected points. Its

hash value is calculated With the salt of the account and
compared With the stored hash value to decide if access is
granted or not.

outside contour of a character are not used since those points

might be occluded or inside the area overlapping With the

neighboring character.
[0114]

The primitive domain of TextPoints may consist of

internal and invariant cross-points of the character set used to

generate TextPoints images. The ?ve cross points (X) 1302

shoWn in FIG. 13A meet the requirements. But the points
1302a and 1302b, and similarly the points 13020 and 1302d,
are very close. For some alloWed fonts of A, these tWo

points might be Within the alloWed select error that it is hard

to tell What the users intention is. Therefore, the distance of
any tWo points in the primitive domain belonging to the same
character must be larger than a preset threshold. A possible
selection of the primitive domain for letter A is the three
cross points Q() 1304, 1304a, and 1304b shoWn in FIG. 13B.
Characters Without any internal cross-point, e.g., letter S,
do not contribute any points to the primitive domain of Tex
tPoints. These characters can still be used in generating CaRP
images. They are cluttering characters to confuse computers.

[0115] It is Worth comparing potential passWord points

betWeen TextPoints and existing select-based ?ne granularity
graphical passWords such as PassPoints. In PassPoints,
salient structural points should be avoided since they are
readily picked up by an adversary to mount a dictionary
attack. Avoiding these structural points makes it hard for
humans to remember the points in a passWord for a long time.

Variations and Discussion

[0120] CaRP schemes Working With the HTTP Basic

Access Authentication that employs the transport layer secu
rity HTTPS to secure the communications betWeen the Web
server and a client during a user authentication have been

described above. HoWever, the HTTP standards also support

the HTTP Signature Access Authentication, Which is a chal

lenge-response authentication protocol. This protocol is

rarely used in practice since it still needs HTTPS to provide
secure communications betWeen the client and the server.

Nevertheless, a CaRP scheme, called TextPoints for Chal

lenge-Response (TextPoints4CR) CaRP is described beloW.

[0121] A CaRP image is a CAPTCHA image. Visual
objects in a CaRP image are typically Warped or deliberately
distorted to prevent computers from recognizing them. In
other Words, passWord-element indicators are typically
Warped or distorted versions of potential passWord elements.
This may have adverse impact on user experiences since
humans may need a longer time in recognizing the visual
objects in a CaRP image than the case When the visual objects
are not Warped or distorted. It is possible to combine CaRP
With a soft keyboard or other Ways to enter a passWord (e. g.,

typing a passWord With a keyboard When characters are used):

at one time a CaRP image is used, and another time a soft

keyboard is used. A soft keyboard is an image Which

These con?icting requirements have forced existing select

displays potential passWord elements. Potential passWord

based ?ne granularity graphical passWords to strike a tradeoff

betWeen security and usability. TextPoints addresses this

elements are easier for humans to identify than passWord

element indicators since passWord-element indicators are

problem by removing the con?ict. In TextPoints, structural

typically a Warped version of potential passWord elements. In

points are used to help remember the points selected in a

passWord. They are dynamic (as compared to static points in

existing graphical passWord systems) and contextual:

[0116]

1. Dynamic: The locations of the points in the

some embodiments, different incarnations of a potential pass

Word element may be used in different soft keyboard

instances or images. For example, at one instance of a soft
keyboard, letter A appears as Times NeW Roman font. In
another instance of a soft keyboard, letter A appears as

primitive domain as Well as their contexts (i.e., charac

ters) vary from one image to another. The points in one
image are computationally uncorrelated to the points in

ments in a soft keyboard are arranged in a random order. In

another image.

another embodiment, the potential passWord elements in a

[0117]

Arial font. In one embodiment, the potential passWord ele

2. Contextual: Whether a similarly structural

soft keyboar are arranged in a preset order. For a CaRP

point is in the primitive domain (thus a candidate point

using characters as potential passWord elements, the soft

of passWords) or not depends on its context. It is only if

keyboard is a conventional soft keyboard. For a CaRP using

animals as potential passWord elements, a soft keyboard
may be an image containing different animals.
[0122] Different mechanisms can be applied to determine

Within the right context, i.e., at the right location of the

character it belongs to.
[0118] This context-dependence requires an adversary to
recogniZe the correct contexts, i.e., characters, ?rst. If this

When a CaRP is used and When a soft keyboard is used. In

Apr. 18, 2013

US 2013/0097697 A1

some embodiments, a decision mechanism similar to that

selected point is said to be the point associated With the

used in CAPTCHA-based Password Authentication (CPA)

boundary point. Once having processed all the points in the

primitive domain, a Warping operation is applied on the image

protocols can be applied: When CPA decides a CAPTCHA

or a computing system 202 in FIG. 2 or portable user device

302 in FIG. 3 is used, and a soft keyboard is used When a
knoWn device is used. A device is knoWn if the user has a

to move all the boundary points in the primitive domain to

their associated points While all the internal points in the
primitive domain stay at the same location. On the resulting
image, all the points in the primitive domain are internal
points, each at least '5 distance aWay from the boundary of its
discretiZation grid-square. The image is then sent to the client
to select her passWord. If the users select error is Within '5,
then each selected point falls into the same discretiZation

image should be used, a CaRP image is used; and When CPA

decides a CAPTCHA image should not be used, a soft key
board is used.
[0123] In one embodiment, a CaRP image is used When an
unknown device such as a user device 10411 or 10419 in FIG. 1

successful login Within a ?xed period of time. In some

grid-square as the original passWord point. Therefore identi

embodiments, a device has also to be a trusted device to be a

knoWn device. In one embodiment, a knoWn device may be

cal sequences of discretiZed grid-squares are generated: the

identi?ed by its IP address or device ID stored With the user

ID in a user pro?le database 108 in FIG. 1 oruser pro?lers 219
or 314 in FIGS. 2 and 3. In another embodiment, a knoWn
device may be identi?ed by a cookie stored on the device. A
CaRP image may be used When a user fails in a number of

passWord retrieved after the userID is received, and the client

login attempts With a knoWn device.

TextPoints4CR requires the authentication server to generate

[0124]

In another embodiment, soft keyboard is used

unless a user has failed in a number of lo gin attempts, and then

a CaRP image is used. Different thresholds of failed login

attempts may be applied, depending on if the device is knoWn
or unknoWn. A small threshold of failed login attempts (e.g.,
3) may be applied When a device is unknoWn and a large

authentication server generates the sequence from the stored

generates the sequence from the user selects. These tWo iden
tical sequences are then used as if a conventional secret shared

betWeen the tWo parties in challenge-response authentication.

[0128] Unlike

other

proposed

CaRP

schemes,

the same sequence of discretiZed grid-squares as a user cor

rectly selects during a user authentication Without relying on

any information from the client. Storage of passWords at an
authentication server might be a security concern since an

inside attacker might be able to steal. To mitigate this prob

lem, user passWords should be encrypted by a master key that

threshold of failed login attempts (e.g., 30) may be applied

only the authentication server knoWs. During a user authen

When a device is knoWn.

[0125] In some embodiments, another Way to enter a pass

tication, the encrypted passWord is retrieved using the

Word other than using a soft keyboard is used With CaRP

received userID, decrypted, and then used to generate the

sequence of discretiZed grid-squares. The decrypted pass

images. For example, When characters are used, the soft key

Word is then deleted from memory.

board is replaced With a real keyboard that a user enters her

passWord by typing on the real keyboard.

Illustrative Computing Device

[0126] TextPoints4CR is a CaRP modi?ed from the Text

Points for a general challenge-response user authentication.

[0129] FIG. 14 shoWs an illustrative computing device

1400 that may be used to implement the security administra
tor 102 and/or the security module 214/310. It Will readily be
appreciated that the various embodiments described above

Unlike TextPoints Wherein the authentication server stores a

may be implemented in other computing devices, systems,

salt and the hash value of the passWord and the salt along With

and environments. The computing device 1400 shoWn in FIG.

14 is only one example of a computing device and is not

TextPoints4CR CaRP

the userID, the authentication server in TextPoints4CR stores

the passWord and userID. Another difference betWeen these

intended to suggest any limitation as to the scope of use or

tWo CaRPs is that a TextPoints4CR image does not contain

any repeated characters While a character in a TextPoints

functionality of the computer and netWork architectures. The

requires both the server and the client to generate an identical

computing device 1400 is not intended to be interpreted as

having any dependency or requirement relating to any one or
combination of components illustrated in the example com

sequence of discretiZed grids Without exchanging any infor

puting device.

mation about the sequence. Therefore there must be a unique

Way to generate this sequence from the shared secret, i.e.,
pas sWord. If there are repeated characters Which contains the

[0130] In a very basic con?guration, the computing device

1400 typically includes at least one processing unit (or pro
cessor) 1402 and system memory 1404. Depending on the

image can be repeated. This is because that TextPoints4CR

points used in the passWord, generation of this sequence is no

exact con?guration and type of computing device, the system

longer unique. This sequence is used as if the secret in a

procedure to generate a TextPoints image is ?rst applied.

memory 1404 may be volatile (such as RAM), non-volatile

(such as ROM, ?ash memory, etc.) or some combination of
the tWo. The system memory 1404 typically includes an oper
ating system 1406, one or more program modules 1408, and
may include program data 1410. The computing device 1400
is of a very basic con?guration demarcated by a dashed line

Then all the points in the primitive domain are located on the

1414.

image, overlaid With the discretiZation grid. For every point in

[0131] The computing device 1400 may have additional

features or functionality. For example, the computing device
1400 may also include additional data storage devices (re
movable and/or non-removable) such as, for example, mag
netic disks, optical disks, or tape. Such additional storage is

conventional challenge-response authentication protocol.

[0127] Suppose the tolerant select error along both x-axis
and y-axis is t, and the discretiZation grid siZe is ui4'c on both
directions. To generate a TextPoints4CR image, the same

the primitive domain, the distance to the center of the discreti

Zation grid-square it is in along both x-axis and y-axis is

calculated. Apoint is said to be an internal point if the distance
is less than 0.5[1-"5 on both directions; otherWise a boundary

point. For each boundary point, a nearby internal point in the

illustrated in FIG. 14 by removable storage 1416 and non

same discretiZation grid-square is randomly selected. This

removable storage 1418. Computer-readable media may