Loading...

Deanonymization of social network profiles by the use of pictures meta data

Seminar Paper 2011 24 Pages

Computer Science - IT-Security

Excerpt

Contents

Glossary

1 Introduction

2 State of the art
2.1 Picture meta data
2.2 Pattern noise from digital camera
2.3 Modifiability
2.4 Summary

3 Use of profile information
3.1 Textual information
3.2 Picture information
3.3 Summary

4 Detection of relationships

5 Summary
5.1 Strengths of presented techniques
5.2 Weaknesses of presented techniques
5.3 Likelihood of deanonymization
5.4 Outlook

Glossary

illustration not visible in this excerpt

List of Figures

2.1 Extract from EXIF 2.2 Specification
2.2 Extract from EXIF data, HP Photosmart R740
2.3 Extract from EXIF data, IPhone
2.4 Camera Sensor (CCD-Sensor)1
2.5 Camera Body2

3.1 Anonymous StudiVZ profile
3.2 Anonymous Facebook profile

List of Tables

5.1 Strengths of textural data

5.2 Strengths of EXIF data

5.3 Strengths of PRNU

5.4 Weaknesses of textural data

5.5 Weaknesses of EXIF data

5.6 Weaknesses of PRNU

1 Introduction

By the spread of mobile devices and netbooks mobile internet accessibility is getting easier day by day. In parallel to this movement, the number of social network users is growing constantly.1

Today social networks are well integrated in our daily live and accessible with ease. They provide a range of functionality which is used in business as well as private surroundings.

Motivated by this situation and the existence of improvidently users, social networks provide more and more private and business information. One aspect by the growing number of users could be the possibility to scale the level of anonymity, and related to this a fiction of privacy provided by the social networks. Id est internet users can act with there real name and write a blog about there experiences and opinions, but on the other hand they are able to act with a minimum relation to them-self.

In the surrounding of social networks this user behaviour tends, on the good side, to a freedom of expression and a profit on opinion formation, like Twitter and other social networks presented in news.2 But on the objectionable side of anonymity users can easier be discriminated or cyber-bullyed by anonymous others. Largely unknown, however, is that even anonymous users, in some cases, can be identified in the web. This possibility to identify a person by it’s profile can be conflict with it’s interests. Basically if the user does not want to be identified, e.g. if she/he wants to act anonymous in a self-regulating community forum or other social networks in the web. Thereby users decision to act anonymous is not given any more if they publicize profile pictures within meta data and pattern noise. By comparing of this meta data with multiple social network user profiles the believed anonymity of a user can be fiction.

Modern user profiles in social networks like Facebook, StudiVZ, Xing or LinkedIn differ from each other because of the main purpose the network has build around. This fact makes an overall user profile identification more difficult and reduces the rate of useful information. Nevertheless actual movements in computer security and forensic science show that linking of uncovered relationships between social network profiles is possible.3

2 State of the art

To get a deeper understanding in the techniques of deanonymization, the following chapters show two ways of information generation which are not recognized by an average user. Therewith users don’t recognize the magnitude of published information. Further on I will focus on pictures presented in an user profile and available for others.

2.1 Picture meta data

Modern digital cameras functionality has been growing over the time and the possible picture resolution too. In this case it is not unusual for modern devices to provide a resolution higher then 5MP1. Mobile- and smart phones as well as traditional digital cameras save images generally in formats like JPEG2 or TIFF3. These standardized formats allow to embed meta data, which is practical used to organize pictures after creation and to specify the image for further editing. Embedded meta data in the mentioned picture formats are represented by EXIF4 -meta-data and IPTC5.6 These two format types, to integrate image related information into pictures, are technical equivalent and provide nearly similar functionality.7

EXIF EXIF meta information are stored in the first 64KB of an image. The EXIF standard8 includes a variety of possible information that can be stored in this sector of the image. Figure 2.1 shows a number of embeddable information.

As examples figure 2.2 shows EXIF information of a HP® Photosmart R720 digital camera and figure 2.3 shows EXIF information of an IPhone. As we can see from the picture shown, basic information about the image (size, exposure time and date) and the device (manufacturer and model) can be concluded from meta data. In the sense of data privacy some of these information are individual-related data. The GPS9 -tag in figure 2.3, for example, indicates the position where the image was made. If the pool of image data is bigger then one picture, it will be possible to convey geographical points. By the combination of all available information of a social network profile - name, profession, company and so on -, geographical information can be related to specific activities, events or simply locations, or at least to identify or classify other pictures available in the web and social networks.

illustration not visible in this excerpt

Figure 2.1: Extract from EXIF 2.2 Specification

This available information, witch is accessible with ease by professional photo ed- itors, can be adapted and supplemented. Therewith customized EXIF-tags can be added to include individual information like copyrights or names of persons shown on the picture. However, concluded by the level of adaptability meta data can as well be basically copied from one picture and added to an other, to hide the original information or to add meta data later on.

illustration not visible in this excerpt

Figure 2.2: Extract from EXIF data, HP Photosmart R740

illustration not visible in this excerpt

Figure 2.3: Extract from EXIF data, IPhone

2.2 Pattern noise from digital camera

An other methodology to identify social network profiles by the use of pictures in- formation are given by the unique pattern noise that is part of every digital cameras image. The overall picture noise is represented by a sum of different noise types that are by itself depending on different physical attributes. The main noise components are pattern noise, temperature-dependent noise and random physical impacts.

Pattern noise The pattern noise represents the most important component for de- sanonymization. This camera sensor noise is statistically detectable and is assumed to be normally distributed. It is mainly composed from dark current and pixel to pixel inhomogeneities (PRNU10 ). The reason for this is the sensors production pro- cess. During the production of a digital camera sensor small manufacturing errors occur and result in slightly differences in the sensors optical property. As figure 2.4 illustrates, lateral overflow of sensor material represents one way PRNU is generated. The resulting errors are not related to the production method of camera sensors as such, more to the fact that every production process is subject of specific tolerances and conditions.

illustration not visible in this excerpt

Figure 2.4: Camera Sensor (CCD-Sensor)11

To obtain the noise pattern, the following mathematical transformations are neces- sary. In the below shows equations, let X be a picture to be examined and let F be the filter to extract the pattern noise. It follows, that X − F (X) would be defined as the equation to extract the pattern noise. If F is chosen as a wavelet denoise filter or Gaussian filter the outcome will be from respectable result. If further more c would be the correlation coefficient between the extracted pattern noise (X − F (X)) and the reference pattern P i of an camera C i, the equations below represent the relation between camera and the image that is examined. Like Luk (2006) explains, existence of a camera reference pattern P i is not necessary required, if enough images exist, which can be assigned to a specific camera. Luk (2006) and Löhr (2010) used the average of 20 up to more then 50 pattern noises to generate the reference pattern P i,

illustration not visible in this excerpt

Temperature noise In contrast to the described pattern noise, noise generated by the temperature is coming into existence by non-uniform temperature zones on the camera sensor (s. figure 2.4). Like figure 2.5 of an Olympus camera shows, cameras body is connected to the sensor and consequently thermal energy is transmitted from the body to the sensor. This effect is aided by the body material and its thermal conductivity, specially if the camera is carried in hand rather then placed on a tripod, or placed in sun.

Another difference between pattern noise and temperature noise is the fact that temperature noise is a dynamical component of the overall noise represented in an image. Therewith can be concluded that temperature fluctuations has a negative impact on pattern noise.

illustration not visible in this excerpt

Figure 2.5: Camera Body13

Other physical impacts As mentioned at the beginning of the chapter, more like the two presented noise types exist. Depending on the physical environment the camera is exposed, the proportion of different noise pattern vary. Dirt or scratches are pos- sible influences in this context, which as well lead in characteristic noises.

[...]


1 cf. Schmidt (2010) and lis (2010)

2 cf. Neumann (2011)

3 cf. Greveler u. Löhr (2010) and Geradts u. a. (2009)

1 Mega Pixel (one million pixels), Used as a term to represent the resolution of camera sensors and displays (e.g: 1280x800 Pixel = 1MP)

2 Joint Photographic Experts Group, file interchange format (ISO/IEC 10918)

3 Tagged Image File Format

4 Exchangeable Image File Format

5 International Press Telecommunications Council

6 cf. Electronics u. Association (2010), (IPTC)

7 cf. Greveler u. Löhr (2010)

8 Electronics u. Association (2002), Electronics u. Association (2010)

9 Global Positioning System

10 Photo-Response Non-Uniformity noise

11 cf. Löhr (2010), 28/02/2011

12 cf. Kirchner (2006), Löhr (2010) and Che (2007)

13 Source: http://www.imaging-resource.com/PRODS/E20/ZHEATSINK.JPG, 28/02/2011

Details

Pages
24
Year
2011
File size
826 KB
Language
English
Catalog Number
v171059
Institution / College
University of Applied Sciences Münster
Grade
1.3
Tags
Picture meta data meta data EXIF noise pattern PRNU Photo-response Non-Uniformity noise scocial network social network profile deanonymization user identification

Author

Share

Previous

Title: Deanonymization of social network profiles by the use of pictures meta data