Genetically optimized
face image CAPTCHA
Guided
by
BINDU
ELIAS
Presented by
ROMY GEORGE
S7 EB, 52
1
EEE DEPT. MACE
Genetically optimized face image CAPTCHA
INTRODUCTION
Test Chart
WHAT IS CAPTCHA?
Completely Automated Public Turing Test to Tell Computers and
Humans Apart

Differentiate between humans and bots.

Give tasks easier for humans but difficult for bots to complete.
2
EEE DEPT. MACE
EEE DEPT. MACE
Genetically optimized face image CAPTCHA
Why CAPTCHA?




3
Reduce e-mail spams.
Stop automated blog and forum responses.
Prevent denial of service (DoS) attacks on web servers.
Prevent bots from taking part in online polls, registering for free email accounts and collecting e-mail addresses.
EEE DEPT. MACE
EEE DEPT. MACE
Genetically optimized face image CAPTCHA
Existing CAPTCHAS!
1.
2.
3.
4.
5.
4
Text based
Generic image based
Speciality image based
Knowledge based
Audio based
EEE DEPT. MACE
EEE DEPT. MACE
Genetically optimized face image CAPTCHA
Disadvantages of existing CAPTCHA
Language dependent and not suitable for
multilingual world wide usage.
Cannot be used effectively in mobile touch
platforms.
More vulnerable to attack.
5
EEE DEPT. MACE
EEE DEPT. MACE
Genetically optimized face image CAPTCHA
Face image CAPTCHA : fgCAPTCHA

Generation of computationally
challenging face detection
CAPTCHA.
Complex
background
generation
Face and nonface image
selection

Utilization of GENETIC
ALGORITHM to optimize
CAPTCHA parameters.
Distortion
selection
Initial
chromosome
generation
6
EEE DEPT. MACE
EEE DEPT. MACE
replacements
Termination
criteria
mutation
Face image
CAPTCHA
crossover
selection
Distortion optimization using genetic algorithm
Genetically optimized face image CAPTCHA
The generation process can be represented by
C∅ is a CAPTCHA with distortion ∅ applied.
When a simulation is used to get results, we have:
7
EEE DEPT. MACE
EEE DEPT. MACE
F is the fitness
value, indicating
difference between
two likelihoods
Genetically optimized face image CAPTCHA
Background generation and image selection
• a 400x300 pixel background
of overlapping rectangles of
different colors and sizes is
generated.
• Colored rectangles are
scattered across such that
95% of background is
covered.
8
EEE DEPT. MACE
EEE DEPT. MACE
• Minimum two face images
are selected.
• At least one non face image
is also selected.
Genetically optimized face image CAPTCHA
DISTORTION SELECTION
9
EEE DEPT. MACE
EEE DEPT. MACE
Genetically optimized face image CAPTCHA
Distortion optimization using genetic algorithm


We need to find the distortion
settings which generate the
optimized CAPTCHA.
We use genetic algorithm to
efficiently identify optimal
distortion setting.
Fig: Image formed before distortion is applied.
10
EEE DEPT. MACE
EEE DEPT. MACE
Genetically optimized face image CAPTCHA
GENETIC ALGORITHM


11
A genetic algorithm (or GA) is a search technique used in
computing to find true or approximate solutions to optimization
and search problems.
Use techniques inspired by evolutionary biology such as
inheritance, mutation, selection, and crossover (also called
recombination).

Implemented as a computer simulation.

Solutions can be represented in binary.
EEE DEPT. MACE
EEE DEPT. MACE
Genetically optimized face image CAPTCHA
Steps in genetic algorithm
initial population
selection
Crossover
mutation
Run replacements
No
terminate
12
EEE DEPT. MACE
EEE DEPT. MACE
Yes
Genetically optimized face image CAPTCHA
1.Generation of initial chromosomes




13
A set of 150 chromosomes are
selected.
Each chromosome represent
one combination of distortion
settings.
a chromosome contain two
genes each encoding a
distortion type and its real
valued intensity.
A fitness value is calculated for
each chromosome using
equation 3.
EEE DEPT. MACE
EEE DEPT. MACE
Genetically optimized face image CAPTCHA
2. Selection of candidates for next generation.


14
A roulette wheel based process is used to
select chromosomes.
The process select chromosomes at a rate
proportional to their fitness.
EEE DEPT. MACE
EEE DEPT. MACE
Genetically optimized face image CAPTCHA
3. Perform crossover


15
Crossover is best understood
visually.
Single point crossover.
EEE DEPT. MACE
EEE DEPT. MACE
Genetically optimized face image CAPTCHA
4. Perform mutation

16
Mutation maintains genetic diversity from one
generation to the next.

alters one or more gene value from its initial state.

It preserves and introduces diversity.
EEE DEPT. MACE
EEE DEPT. MACE
Genetically optimized face image CAPTCHA
5. Run replacements and evaluate termination
criteria
Running replacements: chromosomes with the best fitness value is
kept from both parent and child generations.
generations
Time limit
Termination
criteria
17
EEE DEPT. MACE
EEE DEPT. MACE
Stall
generations
Fitness limit
Genetically optimized face image CAPTCHA
Examples of face image CAPTCHA
18
EEE DEPT. MACE
EEE DEPT. MACE
Genetically optimized face image CAPTCHA
Comparison of fgCAPTCHA with other powerful
CAPTCHAS
Comparison of fgCAPTCHA with reCAPTCHA and IMAGINATION yield the
following results on mobile devices:
• Human success rate is maximum for reCAPTCHA and fgCAPTCHA.
• automated attack success
rate is limited for
100
fgCAPTCHA.
80
• fgCAPTCHA is more user
friendly.
60
fgCAPTCHA
40
reCAPTCHA
20
IMAGINATION
0
19
EEE DEPT. MACE
EEE DEPT. MACE
human
success
rate
automated user ranked
attack
easiest to
success
use
Genetically optimized face image CAPTCHA
CONCLUSION




20
Face image CAPTCHA incorporates improved visual distortions which
strengthens the security.
Remove dependency on humans for parameter selection and
optimization.
Uses GA based image generation which increases human success rates
and reduces automated attack rates.
Achieve 88% accuracy rate during evaluation for humans.
EEE DEPT. MACE
EEE DEPT. MACE
Genetically optimized face image CAPTCHA
REFERENCES
1
fgCAPTCHA: Genetically Optimized Face Image CAPTCHA ,BRIAN M.
POWELL1, GAURAV GOSWAMI2,IEEE Access ,date of publication April
29, 2014, date of current version May 21, 2014.
2
R. Datta, J. Li, and J. Z. Wang, ``Exploiting the human-machine gap in image
recognition for designing CAPTCHAs,'' IEEE Trans. Inf. Forensic Security, vol.
4, no. 3, pp. 504518, Sep. 2009.
3
21
M. Frank, R. Biedert, E. Ma, I. Martinovic, and D. Song, ``Touchalytics: On
the applicability of touchscreen input as a behavioural biometric for
authentication,'' IEEE Trans. Inf. Forensics Security, vol. 8, no. 1, , Jan. 2013
4
D.-J. Kim, K.-W. Chung, and K.-S. Hong, ``Person authentication using
face, teeth and voice modalities for mobile device security,'' IEEE Trans.
Consum. Electron., vol. 56, no. 4, pp. 26782685, Nov. 2010.
5
H. Lee, S.-H. Lee, T. Kim, and H. Bahn, ``Secure user identication for
consumer electronics devices,'' IEEE Trans. Consum. Electron., vol. 54,
no. 4, pp. 17981802, Nov. 2008.
EEE DEPT. MACE
EEE DEPT. MACE
Genetically optimized face image CAPTCHA
U
Any Questions ?
22
EEE DEPT. MACE
EEE DEPT. MACE