Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ......

91
Privacy in Public, and Security without Barriers Prasant Mohapatra University of California, Davis

Transcript of Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ......

Page 1: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Privacy in Public, and Security without Barriers

Prasant MohapatraUniversity of California, Davis

Page 2: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Privacy in Public• Privacy: Identity and all related information• Public: Physical space and cyber space• Privacy in public space, privacy in social networks,

privacy in cyberspace – sounds like an oxymoron!• Consequence: Identity theft, surveillance, information

overload, targeted advertisements• Although highly desirable to have privacy in public, the

first step should be the awareness and quantification of privacy leak

• Privacy in public: Can we maintain some level of privacy while in public?

2

Page 3: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Security without Barriers• Although desirable, security impacts performance and

usage convenience• Imposes several barriers – quality of experience,

performance, utility• Exploiting the performance barriers as signatures for

facilitating security - forensics• Simplifying authentication while maintaining robustness• Hard to enforce security without barriers – can we

minimize or obfuscate those?

3

Page 4: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Organization

• Privacy in Publico Privacy leakageo Privacy Preserving Tracking

• Security without Barrierso Secret Message Sharing in Social Networkso Live Video Forensicso Sensor Assisted Authentication

• Concluding Remarks

4

Page 5: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Characterizing Privacy Leakage

Motivation:

o What is leaked in public?o Quantifying privacy leakageo Enhance user awareness of privacy protectiono Improve website/app privacy designo Improve network protocol design

5

Page 6: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Privacy leakage sources

• Client devices• Website-browsing content• Profiled ads from third party advertiser• A combination of these sources reveals a

significant amount of private information

6

Page 7: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Our Methodology• Look at DNS packets, Probe Request Frames from user

deviceso For user name information o Location informationo Other information

• Concatenating the “host”, “directory” and “filename” in the HTTP header fields.o For website urlso For website contento For ad content

• Count the pieces of privacy information (privacy units) leaked in different privacy categories

7

Page 8: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Privacy Leak from User Device• Device nickname

o Many users set device nicknames as their own names o Example: device nick name in a Multicast DNS packet

Domain Name System (query)Flags: 0x0000 (Standard query)QueriesGINGERs-iPhone.local: type ANY, class IN, "QU"

• Broadcasting SSID listo SSID list reveals the user’s previous accessed networkso Example: SSID in a Probe Request Frame

IEEE 802.11 wireless LAN management frameTag: SSID parameter set: UC Davis

8

Page 9: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Privacy Leak from Web Accesses• Privacy info source

o URL linko Content in the website

• Possible inferenceo Home countryo Hobbyo Locationo Interest merchandizeo Age rangeo Gendero Other

9

Page 10: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

An example:

GET/plugins/activity.php?site=www.cnn.com&action&width=210&height=190&header=false&colorscheme=light&linktarget=_blank&border_color=white&font&recommendations=true HTTP/1.1\r\nReferer: http://www.cnn.com/\r\nAccept-Language: en-US\r\nUser-Agent: Mozilla/4.0 (compatible; MSIE 8.0; Windows NT 6.1; WOW64; Trident/4.0; GTB7.3; SLCC2; .NET CLR 2.0.50727; .NET CLR 3.5.30729; .NET CLR 3.0.30729; Media Center PC 6.0; .NET4.0C; AskTbORJ/5.14.1.20007; MALC)\r\nHost: www.facebook.com\r\n

HTT

P re

ques

t

10

Page 11: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Leaked Information

• The content:

o Ballot Measure Race - 2012 Election Center - Elections & Politics from

o people recommended this.

o Maryland, Maine, Washington approve same-sex marriage; 2 states legalize pot

o 2,770 people recommended this.

o Jay-Z: '99 problems but Mitt ain't one' - CNN.com Videoo 1,717 people recommended this.

• This website content infers user’s location and political interest.

11

Page 12: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Privacy leak from Profiled Advertising

o Most of the ads pushed to the user are based on profiling from third party advertisers

o Possible inferences of ads:o Interest in merchandizeo Hobbyo Age rangeo Gendero Locationo other

12

Page 13: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

An example:

GET /pagead/ads?client=ca-pub-7712942546971716&output=html&h=200&slotname=1003552717&w=200&flash=11.1.111&url=h ttp%3A%2F%2Fwww.uzivoradio.com%2Fs-beograd.html&dt=1336765628611&shv=r20120502&js v=r20110914&saldr=1&prev_slotnames=0178323763&correlator=1336765558769&frm=20&adk =174632003&ga_vid=1235498640.1336765628&ga_sid=1336765628&ga_hid=1546525153&ga_fc =0&u_tz=-240&u_his=4&u_java=0&u_h=1347&u_w=1004&u_ah=1347&u_aw=1004&u_cd=32&u_npl ug=1&u_nmime=2&dff=sans-serif&dfs=16&adx=0&ady=286&biw=1004&bih=1347&oid=3&ref=http%3A%2F%2Fwww.uzivoradio.com%2Findex.php%3Fstrana%3Dprivacy&fu=0&ifi=2&dtd=M&xpc =bvF8Tksvg2&p=http%3A//www.uzivoradio.com HTTP/1.1 Host: googleads.g.doubleclick.net Accept-Encoding: gzip Referer

HTT

P re

ques

t

13

Page 14: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Privacy LeakContent:

Hr Service Custom Solution/Free Consultations Experienced Consultants-Call Today.www.InfinitiHR.comBelmont TV - Since 1943 3D, LED, LCD, & Plasma TV Sale. Free Delivery & Setup. 3 Locations.www.BelmontTV.com

Loc: Olney, MARYLAND

Loc: ARLINGTON,VIRGINIA

LAUREL, MARYLAND

WHEATON, MARYLAND

ADAD

14

Page 15: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Evaluation

• Real data from 40 airports in multiple countries

• We evaluate the privacy leakage of travelers by pieces of information (privacy unit) concerning their identity, location, social relationship, financial condition and other personal information

15

Page 16: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Results: Privacy Leakage• Photo: website (facebook)• Hobby: website• Name: device name, social website (facebook,

amazon, etc), other website content, apps• Home country: email, website• Shopping interest: website• Location: website content, facebook, ads, apps• Gender: facebook• Travel itinerary: website

16

Page 17: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Websites Accessed in Different Datasets

17

Page 18: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Leaked Information

18

Page 19: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Leaked Information in Different Datasets (1)

19

Page 20: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Leaked Information in Different Datasets (2)

20

Page 21: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Third Party Ads in Different Datasets

21

Page 22: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Privacy Leakage Inferences• Privacy leakages are high in public Wi-Fi

hotspots• Sensitive information is exposed from user

devices, website browsing, profiled ads in DNS, HTTP and other network protocols

• We characterize the leakage of travelers privacy based on real world airport datasets

• Our work triggers the alarm to safeguard the privacy leakages in public area

22

Page 23: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Organization

• Privacy in Publico Privacy leakageo Privacy Preserving Tracking

• Security without Barrierso Secret Message Sharing in Social Networkso Live Video Forensicso Sensor Assisted Authentication

• Concluding Remarks

23

Page 24: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

ALPS: Privacy-Preserving Location Tracking

Page 25: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Continuous Location-based Services

• Periodically and automatically records user’s whereabouts: Location trace

• With more coherent spatio-temporal data: o Better analyze user behavioro Better predict user requirement

• Applicationso Macro-scale: traffic monitoring,

urban planningo Micro-scale: trajectory sharing,

digital personal trainer

Top-right: Nike+; Button-right: Google Latitude

25

Page 26: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Greater Risk to User Privacy

26

Page 27: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Location Privacy Preservation Mechanisms (LPPM)

• Aim at protecting location privacyo Location is treated and protected independently

• Degrade the quality of location sampleso Accuracy: perturbation, dummy locationso Precision: spatial cloaking, temporal obfuscation

27

Page 28: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Trace Privacy Preservation Challenges

• Challenge from correlation in location trace

• Challenge from the advancing capability of adversary

• Challenge from diverse privacy preferences of user

28

Page 29: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Correlation in Location Trace• Location samples are correlated Spatially as

well as Temporally• Correlation can be exploited to partially remove

privacy protection• Example: spatial cloaking

29

Page 30: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Adversary with Contextual Information

• Correlation in trace is determined and can be inferred from contexto Geographic context: road network, dead zoneo Mobility context: speed limit, mode of transport

• Map matching techniques

30

Page 31: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Privacy Diversity: Personalized Privacy

• Different parts of a trace may possess very diverse privacy and QoS requirements

• LPPM should not only capture this personalized preference, but also protect it.

31

Page 32: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Objective• Can we design a LPPM that can:

oResist correlation exploiting attack from contextual-aware adversaries

o Support personalized privacyo Take a mobile-centric and distributed

approachoMinimize the energy and communication

overhead

32

Page 33: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Location-based Service Model

GPS

Cellular Mobile User

LBS Provider

Wi-Fi Positioning Eavesdropper

Adversary

LPPM

33

Page 34: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Adversary Model• Goals

o To reconstruct actual trace from obfuscated traceo To extract user’s personalized privacy preference

• Knowledgeo Localization technologies available o Obfuscation algorithmo Mobility patterno Geographic/topology information

• Map-matching attack

34

Page 35: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Our solution: Adaptive Context-aware Perturbation Framework

Localization Technologies

GPS WPS CID ...

Synthesizer

Perturbed Trace

Separation Tier

Choice

Loca

tion

Sam

ple

Conformation Tier

Reconstructor

NR Adversary DL Adversary

HMMAdversary

Evaluator

Distortion Metric

Reconstructed Trace

Actual Trace

Feedback

LBS UpdateTimer

Distortion Score

...

Privacy Profile

35

Page 36: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Two-Tier Perturbation• Separation Tier:

o Inject noise by choosing from various localization means

o Proportional parameter: A probabilistic control knob for adjusting privacy level

• Conformation Tier: o Reintroduce artificial correlation to perturbed trace

according to context constraints

36

Page 37: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Online Evaluation and Feedback• Modular adversary emulator

o Nearest-Road (NR) Adversaryo Distance-Limit (DL) Adversaryo Hidden-Markov-Model (HMM) Adversary

• Resulted distortion score from emulated reconstruction: o Reflects user’s privacy level at the presence of

adversaryo Provides feed back to setting appreciate proportional

parameter

37

Page 38: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Experiment• Implement the scheme on Android platform with three

available localization technologies• Three map-matching adversary emulator for

reconstructor module• Collect real-life driving traces from two places• Distortion-based metric

38

Page 39: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Performance against Reconstruction

39

Page 40: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Performance in Personalized Privacy

Davis Trace Mountain View Trace

40

Page 41: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Visual Comparison

41

Page 42: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Summarizing Privacy Preserving Tracking

• Protecting privacy of location trace poses several new challengeso Contextual-aware map matching adversary exploiting correlation

in traceo User requirement about personalized privacy for trace

• We propose and design a scheme that meets these challengeso Neutralize the advance capability of map-matching adversaryo Achieve trade-off between privacy and QoS that satisfies

personal privacy requirement

42

Page 43: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Organization

• Privacy in Publico Privacy leakageo Privacy Preserving Tracking

• Security without Barrierso Secret Message Sharing in Social Networkso Live Video Forensicso Sensor Assisted Authentication

• Concluding Remarks

43

Page 44: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Background and Motivation Photo-sharing on online social networks (OSNs) has exploded. Steganography can be used to hide secret messages in images

uploaded – but not easy! Chipping away at censorship firewalls

Photo sharing sites often process uploaded images (e.g., resizing) Official specifications not available Interfere with the use of steganography

Need to exchange secret keys to encrypt hidden messages Steganography does not offer perfect secrecy The availability of out-of-band channel may be difficult

44

Page 45: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Goal and Contributions

How can we provide an effective covert channel using images uploaded on online photo sharing sites ? With the goal of answering this question, we make

the following contributions: Understand how hidden data in images is affected due

to processing on online sites Propose a new simple way of embedding information in

photos to preserve the integrity of secret messages Propose an in-band approach for bootstrapping secret

conversations using uploaded images

45

Page 46: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Feasibility of secret embedding in OSNs

In-depth measurement study of photos uploaded on popular sharing sites Google+, Facebook,Twitter, and Flickr Upload and then download of images, examine if the

message is preserved Steganography tools GhostHost

Embedding after the JPEG End-of-Image marker

StegHide Embedding done at the least significant bit (LSB) of pixel values

Outguess & F5 Embedding at LSBs of DCT coefficients JPEG images are represented with a set of coefficients after Discrete Cosine

Transform (DCT)

YASS Embedding at DCT coefficients but in conjunction with error correcting codes 46

Page 47: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Evaluation of steganography tools onphoto sharing sites

Google+: the most generous and accommodates all the steganography tools. Twitter: the next best;

Facebook and Flickr: the least compatible with steganography YASS: image and redundancy dependent

Tool Facebook Twitter Flickr Google+GhostHost X X X √

StegHide X √ X √

OutGuess X √ X √

F5 X √ X √

YASS √* √ √* √

X = Failure; √ = Success; √* = Conditional Success

47

Page 48: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Impact of processing on hidden messages

Google+ Image integrity is preserved Image size limit for no resizing: 2048 pixels by 2048 pixels

Twitter Metadata fields are cleaned up

Comment field, End-of-image marker

Image size limit: 1024 pixels by 768 pixels Facebook & Flickr Metadata removed Changes in pixel values Changes in DCT coefficients

48

Page 49: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Changes in pixel values and DCT coefficients

Distribution of the differences in the pixel values in two colordimensions between original images and after they are uploaded on Facebook

Variations in DCT coefficients with Facebook and Flickr 49

Page 50: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Surviving Facebook and Flickr

Common embedding approach Embedding in the LSBs of DCT coefficients or pixel values Subject to processing changes Error correction code (ECC) overhead is high Reduce secret capacity

Question: Are there locations within an image that remain relatively unaffected? Maximum change in the pixel values is about 30 Maximum change in the DCT coefficients is 1 Embedding at higher significant bits of a DCT coefficient?

50

Page 51: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Embedding in the second least significantbit (2-LSB) in the DCT coefficients

We modify the open source stego tool F5 F5 embeds the secret message bits in the LSBs of non-zero

DCT coefficients. We embed in the second LSB (2-LSB) of these coefficients

Image capacity In typical images (length and width of 1000 pixels):10,000

non-zero coefficients Using 10% of the capacity can hide 125 byte secret

Visual comparison of original image (left), stego-ed with LSB (middle) and with 2-LSB (right) (10% image capacity used).

51

Page 52: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Enabling Private Communication

Threat model Censor can inspect all publicly available content on the OSN,

and can access privately shared content Censor does not manipulate uploaded content Censor has unlimited access to any steganalysis tool OSN users are who they claim to be

Covert channel to circumvent the censor Use cryptography in conjunction with our proposed 2-

LSB steganographic scheme

52

Page 53: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

In-band key exchange with images A user embeds her public key in her profile photo using

steganography Scenario: A and B are friends in Facebook, B wishes to initiate

the secret communication Bootstrapping steps (three-way handshaking): B fetches A’s profile photo and extracts A’s public key KA

pu

B encrypts a request with KApu and embeds in an uploaded

image A decrypts the request signal from B’s image with his private

key KApr and knows B’s intention to communicate

A then obtains B’s public key KBpu from B’s profile.

A sends an acknowledgement (ack) signal and a symmetric key KS to B; the content is encrypted with KB

pu

B sends another ack encrypted with KS

After the bootstrapping, all the secret messages exchanged between A and B is encrypted using KS 53

Page 54: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Organization

• Privacy in Publico Privacy leakageo Privacy Preserving Tracking

• Security without Barrierso Secret Message Sharing in Social Networkso Live Video Forensicso Sensor Assisted Authentication

• Concluding Remarks

54

Page 55: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Background• Video forensics

o Source identificationo Forgery detectiono Hidden information detection

• Source identificationo Video used as evidence in a court of lawo Track down piracy crimeso Regulate individual video sources

55

Page 56: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Existing Work & Motivation• Existing method

o Watermarko Defective pixelso Sensor pattern noise

• Wireless cameras become very popularo Easy to deploy, especially for large buildings, big

companies, non-professional userso Security camera, surveillance camera

Performance degrades greatly for wireless videos

56

Page 57: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Extract Sensor Pattern Noise

• What is sensor pattern noiseo Non-uniformity of each pixel’s sensitivity to lighto Signature of camera sensor

• Extraction stepso Extract all the noise from a frame

• Assume a frame is a mixture of a locally stationary i.i.d. signal with zero mean and a stationary white Gaussian noise

• Wiener filter o Averaging

57

Page 58: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Extract Sensor Pattern Noise

Add the noises from many frames together:

58

Page 59: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

The Problem• Blurring and blocking

• Why blocking affects traditional technique?o No details in blocks, weaken the averaging resulto Introducing “grid artifacts”

59

Page 60: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Blocking Detection• Existing blocking detection method

o Based on boundary detection and Fourier transformo Time consuming

• Our methodo Step 1: wavelet transform (already done in noise extraction)o Step 2: add results row by row (or column by column)

60

Page 61: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Improved Pattern Noise Extraction

• Exclude the blocking areas• Add up clean frames and clean areas of

blocking frames to calculate sensor pattern noise

• Compare the pattern noise of a video (Nv) with the pattern noise of a camera (Nc)

( )( )( , ) v cv cv c

v cv c

corr − −=

− −

N N N NN NN N N N

61

Page 62: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Wireless Camera Spoofing Attack

• An attacker compromises a legitimate wireless video camera, and sends fake video to the sink using the victim’s identity

• Expedite our methodo Parallelization

• Wavelet transform, local average estimation, etc.o Selective frame processing

• Give priority in extracting pattern noise from I-frames. o Combination of wireless fingerprints

• Packet loss ratio, jitter, average signal strength, signal strength variance, and ratio of blocking frames

62

Page 63: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Experiment Settings• Device

o 7 wireless cameras• 4 × Linksys WVC80N• 1 × Dlink 942L• 1 × Axis M1011w• 1 × Lenovo X301 laptop webcam

o Cisco WRT160N v2 wireless router

• MPEG 4 and 802.11n

63

Page 64: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Performance of Block Detection

Page 65: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

corr of Sensor Pattern Noise

Page 66: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Source Identification Accuracy

Page 67: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Summarizing this part …• We developed a fast yet reliable video

blocking detection method• We developed a source identification

method which works well for wirelessly streamed videos

• We largely improved the source identification speedo Fast enough for wireless camera spoofing attack detection

67

Page 68: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Organization

• Privacy in Publico Privacy leakageo Privacy Preserving Tracking

• Security without Barrierso Secret Message Sharing in Social Networkso Live Video Forensicso Sensor Assisted Authentication

• Concluding Remarks

68

Page 69: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Authentication in Smartphones• Authentication in smartphones

o device unlocko app logino forum/website login

• Authentication typeso Credential-based (User name / password)

• What the user knows• Identity theft• Memory burden

69

Page 70: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Biometric Authentication Voice

• Inconvenient, vulnerable• Requires speaking, Background noise

Fingerprint• Convenient, vulnerable• Expensive hardware required• Limited market

Face (and Iris)• Convenient, vulnerable• Inexpensive – Use mobile camera

Compelling. Let’s explore further

70

Page 71: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Facial Authentication• Face verification / face identification • Face recognition accuracy has been largely

improvedo Accuracy is very close to 100% o Even used for commercial payment systems

• Most smartphones have front-facing cameras; usually higher than 1M pixelso Convenient for face capturingo Quality is good enough for face recognition

71

Page 72: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Facial Recognition• Initially designed to “recognize” not

“authorize” • Surveillance cameras vs. Mobile Device

cameras• Facial biometric software on mobile device

cameras can be spoofed

Type Method

2D photo attack Pictures of a picture, social engineering

2D video attack Video playback, Spoof “Blink” detection

Virtual CameraSoftware

-Advanced Editing Capabilities-Playback can spoof webcam

What if obstacles are removed?

72

Page 73: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Current Status• Android: face unlock alternative since 4.0

o But not many users are using it

• App and website logino User name / password dominates other methods

• Why facial authentication is not widely used in smartphones?o Privacy concernso Security issues

• 2D media attacks• Virtual camera attacks

o usability

73

Page 74: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

2D Media Attack• Photo attack (print attack)

o Use user’s photo to cheat the authentication system

• Video attacko Starting from Android 4.1, eye-blink is requiredo use video to compromise the system

74

Page 75: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

2D Media Attack (cont.)• 3D facial recognition can defend against this

attacko 3D template matchingo e.g. Toshiba Face Recognition Utility

• Difficult to use • Turning heads towards different directions -> user’s burden

o A trial takes more than 20 seconds -> much longer than entering password

o Even a genuine user may need multiple trials to pass

75

Page 76: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Virtual Camera Attack• Virtual camera software

o Add dynamic effects to webcams, make the video look more beautiful and live chat more interesting

o Now become very powerful: stream a pre-recorded video, make OS believe it is captured by a physical cam in real time

o Most of them are for desktop/tablet, but easy to migrate to smartphones

• Use virtual camera software to hack the facial authentication system

76

Page 77: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Our Method• Achieve high security and usability simultaneously

o Safe for 2D media attackso Safe for virtual camera attackso Much faster than 3D face authentication method (speed is comparable

to credential-based method):~2 sec

• How?o Only need to move the phone in front of face for a short distanceo Utilizing motion sensors in smartphoneso No need to move head and sync with directions

donemovehold 77

Page 78: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Counter 2D Media Attack• Idea

o Nose orientation changes when moving phone horizontally if a real 3D face

78

Page 79: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Nose Angle Detection• Detect nose outline

o Video frame preprocessingo Nose detection (can employ existing method)o Nose outline fitting

• Compare nose outline from two sideso Motion sensors: judge the relative position between face and

smartphone, picking correct frame intelligentlyo Light sensor: auto boost screen brightness if dark, to enhance

luminance (improve nose outline detection) 79

Page 80: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Counter Virtual Camera Attack

• Idea !o If real-time video captured by physical cam, small shakes in

video should be consistent with smartphone’s motion sensor readings

o Pre-recorded videos can be detectedo Assume motion sensor readings are not compromised

80

Page 81: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Motion Vector Correlation• Small motions extracted from the video

• Compare with small shakes extracted from motion sensors

81

Page 82: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Evaluations• Samsung Galaxy Nexus with 1.3M pixel front-facing

camera• Android 4.2.2• Video is 480*720@24fps, chopped to 480*640• Use Haar Cascades in OpenCV to detect face and

nose• Face recognition algorithms are orthogonal to our

method, but for completeness, we do include a PCA (principal component analysis) based facial identification module (also implemented using OpenCV)

82

Page 83: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Accuracy of 2D Media Attack Detection

• 9 volunteers, 180 genuine trials• Take 2 photos and 1 video for each: 180 photo

attacks and 90 video attacks

Accuracy using different edge detectors. Choose prewitt hereinafter

Accuracy under different settings

83

Page 84: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Accuracy of 2D Media Attack Detection (cont.)

Accuracy compared with other state-of-art approaches Accuracy under different illuminance

84

Page 85: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Accuracy of Virtual Camera Attack Detection

Gap between genuine trials and attacks (y axis is correlation between small motions extracted from video and motion sensor readings)

Accuracy of virtual camera attack detection

85

Page 86: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Authentication Time

86

Page 87: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Advantages• Dynamic-to-Static credential matching

o Cannot be intercepted, generated or reverse engineeredo Defends against sending OTP to same device o Defends against device cloning

• Ubiquitous o Any mobile device with a camerao Authenticate to Desktop and Mobile

• Simple (as a selfie)o no additional keystrokes

• Quicko Authenticate within 2 seconds

Device Algorithm

Face

Accelerometer

Video

87

Page 88: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Use Cases• Network and/or App Authentication

o BYOD - Corporations, Healthcare, Utilities, MNOs

• Device Access o Controlled/Issued devices – Government

• Financial Institutionso “High-risk” transactions, dual approval

88

Page 89: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Desktop Access to Enterprise

1

2

34

1. End User enters credentials, What You Know (or just enters Username…i.e. dummy terminal)2. SMS/Push Notification is sent to mobile confirming What You Have and maybe Where You Are3. In response, End User take picture proving Who You Are

a. Picture is confirmed on 4Auth server4. Successful Authentication

4Auth3a

89

Page 90: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

For more information …

• Privacy in Publico Privacy leakage (INFOCOM 2013)o Privacy Preserving Tracking (SECON 2013)

• Security without Barrierso Secret Message Sharing in Social Networks (CNS 2014)o Live Video Forensics (INFOCOM 2014)o Sensor Assisted Authentication (MOBISYS 2014)

• Concluding Remarks

90

Page 91: Privacy in Public, and Security without Barriers · PDF fileQueries. GINGERs-iPhone.local: ... • We characterize the leakage of travelers privacy based on real world airport datasets

Thank you!

91