Breaking an Animated CAPTCHA Scheme

Download Report

Transcript Breaking an Animated CAPTCHA Scheme

Breaking an Animated CAPTCHA
Scheme
Vu Duc Nguyen, Yang-Wai Chow and Willy Susilo
University of Wollongong
About CAPTCHA





2
CAPTCHA: Completely Automated
Public Turing test to Tell Computers
and Humans Apart.
Other name: Human Interaction Proofs
(HIPs).
Easily solvable by humans.
Cannot be solved by current computer
programs.
Existing CAPTCHAs are mainly textbased on a static image.
Vu Duc Nguyen,Yang-Wai Chow and Willy Susilo
Breaking an Animated CAPTCHA Scheme
About CAPTCHA
Most of traditional text-based CAPTCHAs are known to be
vulnerable against attacks.

3
Vu Duc Nguyen,Yang-Wai Chow and Willy Susilo
Breaking an Animated CAPTCHA Scheme
About CAPTCHA

Increase the security: distorting, blurring, rotation the
text, overlaying of visual noise. But

Easy for humans
hard for
computers? Not
guaranteed to
exist.
4
Vu Duc Nguyen,Yang-Wai Chow and Willy Susilo
Breaking an Animated CAPTCHA Scheme
Animated CAPTCHA
Static vs. Animation:
Animated CAPTCHAs has been proposed.
Assumption :
 More usability: Animation makes increasing legibility for
humans.
 More security: Distributing the information required to solve
the CAPTCHA challenge over multiple animation frames.
5
Vu Duc Nguyen,Yang-Wai Chow and Willy Susilo
Breaking an Animated CAPTCHA Scheme
Our questions
Animated CAPTCHAs really provide more security ?
How to break animated CAPTCHA and design secure one ?


6
Vu Duc Nguyen,Yang-Wai Chow and Willy Susilo
Breaking an Animated CAPTCHA Scheme
Breaking HelloCaptcha

Breaking a representative animated CAPTCHAs: HelloCaptcha


CAPTCHA provider: Affects many customer’s web sites if broken.
A variety of 84 different variations of 12 categories.
Flitter
Pop Up
Spread Fade
7
H-Mover
Roller
Spring
Mass Flood
Search light
Swapper
Vu Duc Nguyen,Yang-Wai Chow and Willy Susilo
Breaking an Animated CAPTCHA Scheme
Noisy Mosaic
Smarties
Text Flood
Outline
Breaking HelloCaptcha.



Type Distinction.
Single Image Extraction.




By Pixel Delay Map (PDM).
By Catching Line (CL).
By Color Selection (CS).
Pre-Processing and Character Recognition.
Results and Lessons learned.

8
Vu Duc Nguyen,Yang-Wai Chow and Willy Susilo
Breaking an Animated CAPTCHA Scheme
Breaking HelloCaptcha
Type Distinction:
Number of frames
Number of blank frames
Maximum frame delay
Background color
175
55-107
2
8
4 ms
6ms
RGB(255,255,255)
RGB(92, 31, 92)
Results: Most of 84 types can correctly be distinguished 100%.
9
Vu Duc Nguyen,Yang-Wai Chow and Willy Susilo
Breaking an Animated CAPTCHA Scheme
Pixel Delay Map (PDM)


Feature: To get the human user’s attention, the text characters are displayed
at certain fixed locations for longer periods of time
The PDM is an image resulting from the accumulation of the total amount
of time that a pixel gets displayed in a color that is different from the
background color.
10
Vu Duc Nguyen,Yang-Wai Chow and Willy Susilo
Breaking an Animated CAPTCHA Scheme
Pixel Delay Map (PDM)

PDM and extracted static image.
11
Vu Duc Nguyen,Yang-Wai Chow and Willy Susilo
Breaking an Animated CAPTCHA Scheme
Pixel Delay Map (PDM)

PDM on all frames.

PDMs constructed from consecutive 1/6 of the frames.
12
Vu Duc Nguyen,Yang-Wai Chow and Willy Susilo
Breaking an Animated CAPTCHA Scheme
Catching Line (CL)

Character moving areas.

Selected frames and characters by “Catching line”.
13
Vu Duc Nguyen,Yang-Wai Chow and Willy Susilo
Breaking an Animated CAPTCHA Scheme
Color Selection (CS)

Characters separated based on color.
14
Vu Duc Nguyen,Yang-Wai Chow and Willy Susilo
Breaking an Animated CAPTCHA Scheme
Pre-Processing on extracted single image

Noise removal.

Refine by filling.

Shape removal.
15
Vu Duc Nguyen,Yang-Wai Chow and Willy Susilo
Breaking an Animated CAPTCHA Scheme
Character Recognition by OCR program


Use ABBYY FineReader 11.
Use existing embedded training database and/or own
training set .
16
Vu Duc Nguyen,Yang-Wai Chow and Willy Susilo
Breaking an Animated CAPTCHA Scheme
Experimental Results



8,400 animated CAPTCHA samples were collected from
the HelloCaptcha website.
Accuracy of breaking (i.e. correctly recognizing all
characters in the animated CAPTCHA challenges) ranges
between 16% −100% of the time (wide accepted that
more than 1% of the time is essentially broken).
Attacking time: 4 secs/challenge.
17
Vu Duc Nguyen,Yang-Wai Chow and Willy Susilo
Breaking an Animated CAPTCHA Scheme
Lessons learned

The number of frames.
Frame 45

Frame 82
Delay periods:
40ms
18
40ms
Vu Duc Nguyen,Yang-Wai Chow and Willy Susilo
Breaking an Animated CAPTCHA Scheme
1000ms
Lessons learned

Character positions:

The important information is emphasized by displaying it for
longer. That can be exploited using the PDM method.

PDM was used to break 61 of the 84 different types and
can be affected types from other sources:
19
Vu Duc Nguyen,Yang-Wai Chow and Willy Susilo
Breaking an Animated CAPTCHA Scheme
Lessons learned



Moving direction : Only move or scale in the vertical
direction can vulnerable to attacks.
Use of color or luminance: Less is best
Method of delivery: Gif, Flash or Video ?
20
Vu Duc Nguyen,Yang-Wai Chow and Willy Susilo
Breaking an Animated CAPTCHA Scheme
Question ?
21
Vu Duc Nguyen,Yang-Wai Chow and Willy Susilo
Breaking an Animated CAPTCHA Scheme