Department of Informatics, Aristotle University of Thessaloniki Watermarking: Introduction & Background N. Nikolaidis, I. Pitas Dept. of Informatics Aristotle

Department of Informatics, Aristotle University of Thessaloniki

Watermarking: Watermarking: Introduction & BackgroundIntroduction & Background

N. Nikolaidis, I. Pitas

Dept. of Informatics

Aristotle University of Thessaloniki

{pitas, [email protected]}


Digital media: pros & consDigital media: pros & cons

• Digital media advantages:– High quality– Efficient storage– Copies identical to originals– Ease/speed of manipulation/transmission.

• Digital media disadvantages:– Vulnerability to copyright infringement,

tampering, unauthorized distribution.


Some definitionsSome definitions

• Data (information) hiding: embedding messages in digital content.

• Steganography: concealed communication

– Message existence/content is secret.

• Watermarking: the practice of imperceptibly altering a work to insert a message about the work.

– Special case: Perceptible watermarks


Categories of data hidingCategories of data hiding

Message related

to host data

Message not related to host data

Message existence secret

Steganographic watermarking

Covert communication

Message existence

known

Non-steganographic watermarking

Overt communication

(adapted from the book by Cox, Miller, Bloom)


What content to watermark?What content to watermark?

‘Classical’ media types

– Data types that have been used first in watermarking:

• Images: bitmap (binary, grayscale, color)

• Video: uncompressed, MPEG 2

• Audio: uncompressed, MP3



‘Emerging’ multimedia data types:• 3-D images: (voxel-based)

• 3-D models: polygonal meshes, parametric surfaces (e.g. NURBS)

• Vector graphics, GIS data (contours)

• Animation parameters

• Physical objects



‘Emerging’ data types (continued)• Video: MPEG 4 (audio-visual objects)

• Audio: symbolic description (MIDI files)

• Text (natural text ASCII, binary text)

• Software: source code, binary executables, java byte code

• Numeric data sets: stock market data, scientific data


Performance: requirements & Performance: requirements & metricsmetrics

• Imperceptibility– Watermarked data should be perceptually

identical (or similar) to original data.– Objective perceptual quality metrics are difficult

to be devised.– Subjective imperceptibility tests.

Performance requirements vary widely with the application



• Complexity– Embedding & detection/decoding complexity

should be as low as possible.• Embedding/detection execution time (experimental).• Number of elementary operations (theoretical).• Hardware complexity.

• Cryptographical security– Security should be based on secrecy of keys

rather than secrecy of algorithms.– Keyspace cardinality should deter exhaustive

search.



• Detection performance: ability of the algorithm to declare correctly the presence/absence of a watermark.

– Probability of false alarm, Pfa: probability to detect a watermark in a non-watermarked image or an image hosting a different watermark.

– Probability of false rejection, Pfr: probability not to detect the correct watermark in a watermarked image.

– ROC curve: plot of Pfa versus Pfr.


Detection Performance EvaluationDetection Performance Evaluation


Detection Performance EvaluationDetection Performance Evaluation

Receiver Operating Characteristic curve



• Decoding performance: successful retrieval of the hidden message bits. – Bit Error Rate, BER: mean number of

erroneously decoded bits.– Payload: maximum number of bits that can be

encoded in a fixed amount of data and decoded with a pre-specified BER.



• Robustness: detection & decoding performance should not degrade significantly when data are altered due to common processing operations or hostile attacks. (not applicable to fragile watermarks).– Plots of performance versus attack severity.– Breakdown limit for a certain attack: attack

severity where algorithm performance (Pfr, BER) falls below a certain limit.


Taxonomy: robustnessTaxonomy: robustness

• Robust watermarks– Watermarks can survive common processing or

even hostile attacks.• Fragile watermarks

– Watermarks are destroyed by even the slightest host data modification.

• Semi-fragile watermarks– Selective robustness to a set of legitimate

distortions.


Taxonomy: message lengthTaxonomy: message length

• Zero-bit watermarks– The watermarking system can only answer the

question: – “Does the image contain the watermark W?”– “Is this image mine?”

• Multiple bit watermarks– Answer to questions:– “Is the image watermarked?” and

if YES, the embedded message is decoded.– “Whose is this image?”


Taxonomy: embeddingTaxonomy: embedding

• Informed embedding schemes– Exploit the fact that host data are known during

embedding to improve performance.• Watermarking as communications with side information.

• Perceptual masking, informed message embedding, informed message coding (e.g. Quantization Index Modulation).

• Blind embedding schemes– Host data are considered as noise.

• Watermarking as a classical communications problem.

• Spread spectrum methods.


Taxonomy: detectionTaxonomy: detection

• Non-blind schemes– Original content required during detection.

• Alternative terminology: private / informed detection schemes.

• Robust detection / unsuitable for automated search.

• Blind schemes– Detection using only the watermarked content.

• Alternative terminology: public schemes.

• Suitable for automated search / more vulnerable to attacks.


Taxonomy: other criteriaTaxonomy: other criteria

• Embedding domain:– Images: spatial, DFT (magnitude & phase),

DWT, DCT, Fourier-Mellin,…– Audio: temporal, DFT, cepstrum,…

• Watermark removal ability:– Erasable: exact original data can be obtained

from the watermarked data.– Non-erasable


Watermarking ApplicationsWatermarking Applications

• Broadcast monitoring

– Verification of broadcasting of commercials (proper allocation of airtime).

– Audience metering.– Royalty collection/distribution of copyrighted

material (songs, movies).



• Enhancing legacy systems– Embedding enhancement information while

ensuring backward compatibility.• Embedding stereoscopic information in

conventional TV broadcasts.

– A/D survival• Linking of (scanned) printed media to URLs.

• Identifying songs recorded on a PDA and providing additional info through WWW.



• Copyright Protection (owner identification & proof of ownership)

– Tracking of illegal copies of a work. Use as a proof in legal disputes (?).

– Watermark as copyright notice for informing users on the copyright owner and warning for possible copyright infringements.



• Transaction tracking – Tracking the history of the work. – Identification of traitors that illegally

distributed the work.



• Usage Control

– Control the terms of use of digital content: • Prohibit unauthorized recording (copy control).

– DVD copy protection.

• Deny playback of unauthorized copies (playback control).



• Authentication – tamperproofing

– Verify content authenticity of a work:• Declare work non-authentic even when slightly

altered.

• Tolerate certain manipulations (e.g. compression).

• Provide tampered region localization.


Watermarking & Watermarking & CryptographyCryptography

• Cryptography protects data in transit or stored data (not data under processing/viewing).

• Encrypted data need to be decrypted before use.

• Once decrypted, data are no longer protected.– The analog hole (DVD, pay per view).

• Watermarking can safeguard decrypted data.


Watermarking & DRMWatermarking & DRM

• Digital Rights Management: digital content IPR management & protection.

• Watermarking is a major DRM enabling technology.

• Should be combined and compliment other enabling technologies:– Encryption– IPR databases– Fingerprinting– …


Watermarking vs Watermarking vs FingerprintingFingerprinting

• Watermarking: implies modification of the host content (active method).

• Fingerprinting: extraction of robust and concise content descriptors (passive method).

– No host content modification.

– Alternative terminology: robust hashing.

– Essentially a method for multimedia indexing by content.


Past/Current R&D outputPast/Current R&D output

0

50

100

150

200

250

300

1993 1995 1997 1999 2001

Publications

Watermarking publications per yearsource: Peter Meerwald (personal bibliography)


Current technology offerCurrent technology offer

• Core watermarking competence:– Digimarc: DVD IPR, video broadcast

monitoring, image tracking on the web, links for scanned hardcopies

– Diversification to other sectors

– Verance: verification of ads airplay, audio copy protection and rights management.

– Mediasec: printed documents, image & video authentication

– Signum, DCT, Alpha Tec Ltd, …


Current technology offerCurrent technology offer

• Companies dealing in part with watermarking (indicative list):

– Phillips– Microsoft– IBM– Thales – IBM– …

• Watermarking seems to be an attractive topic!


The futureThe future

Will watermarking live up to its promises?

A. The optimistic side“If we can get to the moon or get to Mars… why can’t we put a little

watermark on our content?”

Michael Eisner, CEO, Walt Disney

Fortune, May 27, 2002

“The most promising technical solution for this so-called “analog hole” appears to be watermarking copy control technology…”

Dr. Craig R. Barrett, President and CEO, Intel Corporation

Senate Judiciary Hearing, March 14th, 2002



B. The not so optimistic side“…there is not yet consensus for adoption of any combination of the

proposed technologies. Accordingly, SDMI is now on hiatus, and intends to re-assess technological advances at some later date”

Secure Digital Music Initiative (SDMI)SDMI Plenary, May 18th, 2001

“The Interim Board of Directors of the DVD Copy Control Association (DVD CCA) ended its term July 31, 2002 without selecting a new embedded data (watermarking) copy protection technology.”

August 12th, 2002



• Watermarking technology advocates might have to limit their ambitions.

• Application scenarios where watermarking can indeed be useful should be defined, other applications might have to be abandoned.

• If technology continues to lag behind business needs & expectations for another 2-5 years the future might not be so bright.


The key questionsThe key questions

• Will watermarking truly survive in hostile environments?

• Will watermarking provide strong cryptographic-like security?

• Are there applications where watermarking is the only way to go?

• Is watermarking useful in benign environments?


• This presentation was partially supported by the IST project CETIMARK financed by the European Union.

AcknowledgementsAcknowledgements

Documents

Department of Informatics, Aristotle University of Thessaloniki Watermarking: Introduction & Background N. Nikolaidis, I. Pitas Dept. of Informatics Aristotle