From Digital Watermarking to Multimedia Forensics: How can they help to detect lost identities in multimedia work?

From Digital Watermarking to Multimedia Forensics:

How can they help to detect lost identities in multimedia work?

Shujun LI (李树钧 )Senior Lecturer

Department of ComputingUniversity of Surrey

http://www.hooklee.com

2

Outline

- Where is University of Surrey?

- “Current” “Solution”: Identities as Metadata- “New” Solutions?

- Active Solution: Digital Watermarking- Passive Solution: Multimedia Forensics- Active Solution: Registering Your Work?

- Expertise of Researchers at Surrey Computing

- Questions and Answers

3

Central London Guildford, Surrey

Highway A3: 33 milesFastest train: 33 mins

4

Guildford Station Surrey Computing

Guildford Railway Station Surrey Computing (by walk):17 mins

5

Dept. Computing @ University of Surrey

Detecting Lost Identities:From Digital Watermarking to Multimedia Forensics

Identities in Multimedia Work

7

Current Solution: Identities as Metadata

- Many media formats support metadata (e.g. Exif, IPTC IIM and XMP) as part of their file headers.

8

Identities as Metadata: Problems

- Location is public Unauthorised removal- Content is public Unauthorised modification

9

Solutions?

- Active Solution: Digital Watermarking- Making metadata dependent on the content itself and

secret Change the metadata = Change the content!- Active = Some information (watermark) needs

embedding into the protected work when/after the multimedia work is created in an additional process.

- Passive Solution: Multimedia Forensics- Discovering information which was NOT actively

embedded into the multimedia work.- Where is the information from? – Left by the capturing

device/creator as part of the natural process.- Multimedia work Device Owner (Author?)

10

Solutions? (continued)

- Active Solution: Register Your Work?- Register your multimedia work at a trusted center (e.g.

one maintained by the IPO) actively.- It does not need to be done by the author, but can be

done by the capturing device (when it connects to the Internet) or some software (when a piece of multimedia work is uploaded to the Web).

- One does not have to register the whole work, but only its (cryptographic and/or perceptual) hash value plus some identification information (e.g. device ID, username).

- This can supplement the other two solutions.

Detecting Lost Identities:From Digital Watermarking to Multimedia Forensics

Active Solution:Digital Watermarking

12

What is a digital watermark?

- A digital watermark is a digital artefact embedded into a cover work which is also a digital artefact.

- It can be a textual message, an image or just a random number (like an ID) represented as a number of bits.

- The process of embedding a digital watermark is called digital watermarking.

- It may be visible or invisible. When it is visible, it mimics watermarks printed on the paper.

13

Identities as watermark

- A typical application of digital watermarking is copyright protection.- The watermark is a copyright claim: “All rights reserved

Shujun Li, University of Surrey 2012”.- The watermark needs to be secured! – Only the owner

knows the embedding/extraction key.- The key can be disclosed to settle an ownership dispute.

14

Conflicting objectives

- Fidelity/Imperceptibility = The watermark should not degrade the perceptual quality to compromise the value of the protected work.

- Security = Surviving malicious attacks- Capacity = More information embedded- Robustness/Fragility = Surviving benign processing

- It is impossible to have an ideal system!

Robustness / Fragility

Fidelity / Imperceptibility

S e c u r i t y

Capacity

15

Security is complicated!

- Watermarked-only attack- The attacker has a number of watermarked works only.

- Known-watermark/message attack- The attacker has access to a number of watermarked works and

the watermark(s) embedded in them.- Known-cover/original attack

- The attacker has access to a number of watermarked works and the original cover work(s).

- Chosen-cover/original attack- The attacker can choose the original cover work(s) in which the

watermark(s) is/are embedded.- Known-detector attack

- The attacker has access to the detector.

16

Security is complicated!

- Unauthorized watermark embedding- The attacker wants to forge a watermark or modify a watermark

already embedded.- Example: to claim the ownership over a work of others.

- Unauthorized watermark removal- The attacker wants to remove the watermark embedded in a

watermarked work.- Example: to remove the watermark/fingerprint in a watermarked

(copyrighted) work.- Unauthorized watermark detection

- The attacker wants to detect the watermark embedded in a watermarked work.

- Example: to disclose the author of a watermarked work.

17

Watermarking by cameras

- Digital cameras capable of embedding digital watermarks do (or did?) exist.- Epson PhotoPC 3000Z/700/750Z/800/800Z- Kodak DC200/DC260/DC290

18

Watermarking by cameras

- Paul Blythe and Jessica Fridrich, “Secure Digital Camera,” in Proceedings of 4th Annual Digital Forensic Research Workshop (DFRWS 2004), Baltimore, August 11-13, 2004- Watermark = Biometric (iris) patterns of the

photographer + Cryptographic hash of the image + Time, Date + …

- The watermarking process is LOSSLESS. The distortion caused by the watermark can be restored without any error.

19

Digital watermarking: problems

- Impossible to have an ideal watermarking scheme.- If the accuracy is not too high, can the results be still used?- Can some (advanced) attacks be ignored?

- Configuring a watermarking scheme is very complicated.- How to balance different objectives?

- Require actions from both vendors and users.- Vendors provide the feature.- Users learn and use it properly.

- Somebody needs to keep the secret key and disclose it when needed.

- …

Detecting Lost Identities:From Digital Watermarking to Multimedia Forensics

Passive Solution:Multimedia Forensics

21

What is multimedia forensics?

- A new research topic developed since late 1990s.- Multimedia forensics = Forensic analysis of

multimedia data and devices.- Source identification = What produced this work?

Was this work computer generated or captured by a device? Who produced this work?

- Forgery detection = Is this work genuine? Which part of this work is genuine?

- Processing history recovery = What has happened on this work? What is the parameter of a step of the processing history?

- …

22

Better than digital watermarking?

- Yes- No need to design an (ideal) algorithm.- No need to configure an algorithm.- No action required for vendors or users.

- No- Only limited information can be extracted.- Accuracy of extracted information could be even lower.- Often require training = No a prior information about

something, nothing can be extracted about it.- Anti-forensics can defeat forensics!

23

Identifying authors

- Forensic (behavioural) biometrics- Like handwritten signatures, intrinsic

features in multimedia work may be linked to its authors’ identities.

- BBC News, New Vincent van Gogh painting identified, 9 November 2013- “The museum said the painting was

authenticated by letters, style and the physical materials used.”

- All could be identified via some multimedia forensic techniques.

24

Identifying devices

- Source identification- Identifying the source device producing a digital

audio/image/video file.- How can this be done without a watermark?

- Different hardware and software components in different models

- Different intrinsic features of individual devices (noise patterns, manufacturing deviations, dust spots and dead points on camera sensor array, …)

25

Linking devices with authors

- When the device of the author can be obtained.- Sometimes only the model matters.

- When some other pieces of multimedia work produced by the same device have their author(s) identified.- So “Register Your Sample Work” is important?

- When the device’s intrinsic features are registered at a trusted party.- So “Register Your Device” is important?

Detecting Lost Identities:From Digital Watermarking to Multimedia Forensics

Expertise of Researchers @Surrey Computing

27

MSF Group @ Surrey Computing

- MSF = Multimedia, Security and Forensics- 6 academics

- Prof Anthony TS Ho (Group Head / HoD)- Prof Jianmin Jiang- Dr Johann A Briffa (Lecturer)- Dr Shujun Li (Senior Lecturer)- Dr Norman Poh (Lecturer)- Dr Stephan Wesemeyer (Tutor)

- 1 postdoctoral fellow- 13 PhD students (including 4 jointly with other

research groups)

28

MSF Group @ Surrey Computing

- Multimedia Security- Digital and Multimedia Forensics- Cyber Security and Privacy- Biometrics and Health Informatics- Multimedia Coding & Visual Quality

Assessment- Digital Media Content Analysis- Big Data Analytics

Detecting Lost Identities:From Digital Watermarking to Multimedia Forensics

Thanks for your attention!

Questions + Answers