1CS 161– 13 November 2006© 2006 Doug TygarCS 161 – Watermarking13 November 2006CS 161 – 13 November 2006© 2006 Doug TygarHow can we mark data?We want to protect data:– Video, sound, music (Digimarc, Intertrust, etc)– Programs (Collberg, Thomborson)– Statistical dataExamples of “traditional” protection methods:– False entries in biographical dictionaries– Copyright notices– Licensing agreements– Secure coprocessors2CS 161 – 13 November 2006© 2006 Doug TygarWatermarkingWatermarking: – include low level bit data that marks information– Either on a per-copy basis or a per-provider basisExample: temperature database– slightly adjust temps to mark uniquely• Store copies of info released – If reused, prove using similarities• But what if adversary changes low-level info?CS 161 – 13 November 2006© 2006 Doug TygarCan watermarking work for data?• It is not clear how applicable watermarking will be for data• A perfect technique (immune against strong tampering) is probably impossible• But some watermarking techniques may be usable• DMCA: removing watermarks is illegal• In this talk, I survey watermarking techniques for photographic data.3CS 161 – 13 November 2006© 2006 Doug TygarMotivation• Intellectual property is important for the Internet• IP (images) are valuable– Costly to create high quality images– Users are attracted by good design• Binary data is trivial to copy• The web is a headache for copyright protection• Many methods for free data exchange• Watermarking is seen as the white knight of copyright protectionCS 161– 13 November 2006© 2006 Doug TygarPart 1: Making Image Watermarks4CS 161 – 13 November 2006© 2006 Doug TygarSecrets of a image watermarking salesman• This slide can transform you into an experienced watermarking salesman• Show two identical images• Claim that one is watermarked• Assert that it’s robust against attacks• Get signature on big $ contractCS 161 – 13 November 2006© 2006 Doug TygarCompanies to apply to• Digimarc• Bluespike• MediaSec• Signafy• Signum (signumtech.com)• ARIS (musicode.com)• Intertrust• But also some of the 2-3 letter companies–IBM–HP–NEC5CS 161 – 13 November 2006© 2006 Doug TygarApplications• Copyright protection– Content owner embeds a secret watermark– Proof of ownership by disclosing the secret key• Fingerprinting– Embed a serial number describing the recipient– Later we can detect which user copied the image• Authentication• Integrity verification– A fragile watermark assures integrity• Content labeling• Rights management• Content protectionCS 161 – 13 November 2006© 2006 Doug TygarWatermarking conceptsrobustfragileWatermarking technique Keysymmetric asymmetricWatermarksvisible invisibleVerifiabilityprivate publicOriginal necessaryyesno6CS 161 – 13 November 2006© 2006 Doug TygarVisible watermarks• Visible watermarks are used in special domains– Vatican library– Swiss paper museum• Issues with visible watermarks– Content producer does not like to degrade the image– Customers don’t appreciate them either– Visible watermarks are easier to remove– Easy to detect for people– But more difficult to detect automaticallyCS 161 – 13 November 2006© 2006 Doug TygarThe watermarking process (private wm)• Embed a watermark• Watermark extractioninsertionKextractionSecret keyWatermarkWW7CS 161 – 13 November 2006© 2006 Doug TygarWatermarking process II• Detecting a watermarkinsertionKSecret keyWatermarkWYes / NoCS 161 – 13 November 2006© 2006 Doug TygarRequirements of invisible watermarks• Robust against tampering (un- & intentional)– Various image transformations (RST)– Image compression– Color requantization– Non-linear transformations (print and scan)• Non-perceptible, hard to detect• Easy to use, exportable, etc.• How can watermarking be possible?– The visual system has very strong “error correction”– An images contains a lot of redundancies– Small changes are undetected– People are used to low image quality (TV, newspaper images)8CS 161 – 13 November 2006© 2006 Doug TygarExample: The NEC watermark• There is no perceptible difference between the original and watermarked image• But the difference image looks interesting• The watermark is present everywhere!CS 161 – 13 November 2006© 2006 Doug TygarEarly aproaches:Spatial Domain Embedding• Original idea: LSB is insignificant• JK-PGS (Jordan-Kutter pretty good signature)– The watermark was embedded directly in the LSB of the pixels of the blue plane in the spatial domain– For robustness, every possibility of rotation, translation, scale was searched• Flaws– Blue plane is insignificant– Least significant bits are unimportant– Possible search space is huge– Not secure against, say, compression• Tirkel, van Schnydel, and Osborne scheme– Embed m-sequences in the LSB of the spatial domain– But also not robust against tampering9CS 161 – 13 November 2006© 2006 Doug TygarSpatial Domain Embedding II• Bender ‘95, Nikolaidis and Pitas ‘96– Randomly divide image into disjunct pixel set A and B• For most images, statistically,– Insertion:• choose k small• A pixels: add k• B pixels: subtract k• Merge A and B to get watermarked image– Detection:• divide image again into A and B set• if x close to 0, then no watermark is present• if x close to N*k, then a watermark is present∑∑≈−pixelsA pixelsBpixelpixel__0∑∑−=pixelsA pixelsBpixelpixelx__CS 161 – 13 November 2006© 2006 Doug TygarTransformation Domain Encoding• An early goal was robustness against JPG compression• Hence, design watermarking into JPG compression• New ideas– Use strong error correction– Spread-spectrum encoding– Embed the mark in the perceptually important regions– Tradeoff robustness vs degradation (artifacts)• Robustness against RST is essential– O’Ruanaidh uses Fourier-Mellin transform to achieve RST invariance– Reed-Solomon error correction– Spread-spectrum encoding– Strong error correction also gives JPG robustness– Does not need the original image for watermark extraction!10CS 161 – 13 November 2006© 2006 Doug TygarSignal Processing Primer• The Fourier transform analyzes image frequencies• Properties of the magnitude spectrum– Translation invariance– Rotation of the image translates to a rotation in the Fourier domain–
View Full Document
Unlocking...