Presentation is loading. Please wait.

Presentation is loading. Please wait.

Robust Mesh Watermarking Emil Praun Hugues Hoppe Adam Finkelstein Emil Praun Hugues Hoppe Adam Finkelstein Princeton University Microsoft Research Princeton.

Similar presentations


Presentation on theme: "Robust Mesh Watermarking Emil Praun Hugues Hoppe Adam Finkelstein Emil Praun Hugues Hoppe Adam Finkelstein Princeton University Microsoft Research Princeton."— Presentation transcript:

1 Robust Mesh Watermarking Emil Praun Hugues Hoppe Adam Finkelstein Emil Praun Hugues Hoppe Adam Finkelstein Princeton University Microsoft Research Princeton University Microsoft Research Princeton University

2 Watermarking Applications Authentication / localization of changesAuthentication / localization of changes Fragile watermarks Ownership protectionOwnership protection Robust watermarks Tracing of distribution channelsTracing of distribution channelsFingerprints Authentication / localization of changesAuthentication / localization of changes Fragile watermarks Ownership protectionOwnership protection Robust watermarks Tracing of distribution channelsTracing of distribution channelsFingerprints

3 Watermarking Applications Authentication / localization of changesAuthentication / localization of changes Fragile watermarks Ownership protectionOwnership protection Robust watermarks Tracing of distribution channelsTracing of distribution channelsFingerprints Authentication / localization of changesAuthentication / localization of changes Fragile watermarks Ownership protectionOwnership protection Robust watermarks Tracing of distribution channelsTracing of distribution channelsFingerprints

4 3.How can Alice prove ownership? (and make Bob pay her a lot of money) 2.Bob sells it as his own. Motivating Scenario 1. Alice creates a 3D shape, and publishes it on the web.

5 Digital Watermarks original document watermark kept secret insertioninsertion Hidden in data! publishedpublished extractionextraction detected watermark suspect document attackattack ?

6 Filtering & smoothingFiltering & smoothing A/D & D/A conversionsA/D & D/A conversions ScalingScaling RotationRotation CroppingCropping Filtering & smoothingFiltering & smoothing A/D & D/A conversionsA/D & D/A conversions ScalingScaling RotationRotation CroppingCropping Incidental Attacks

7 Malicious Attacks Adding noiseAdding noise Adding another watermarkAdding another watermark ResamplingResampling Statistical analysisStatistical analysis Adding noiseAdding noise Adding another watermarkAdding another watermark ResamplingResampling Statistical analysisStatistical analysis

8 Our Goal Watermarking scheme for 3D models: Robust against attacksRobust against attacks Works on arbitrary meshesWorks on arbitrary meshes Preserves original connectivityPreserves original connectivity ImperceptibleImperceptible Watermarking scheme for 3D models: Robust against attacksRobust against attacks Works on arbitrary meshesWorks on arbitrary meshes Preserves original connectivityPreserves original connectivity ImperceptibleImperceptible

9 Previous Watermarking [Cox et al. 97] Introduce spread-spectrum for images [Ohbuchi et al. 98] 3 schemes fragile under resampling [Kanai et al. 98] Requires subdivision connectivity meshes [Benedens 99] Redistributes face normals by moving vertices [Cox et al. 97] Introduce spread-spectrum for images [Ohbuchi et al. 98] 3 schemes fragile under resampling [Kanai et al. 98] Requires subdivision connectivity meshes [Benedens 99] Redistributes face normals by moving vertices

10 Spread-Spectrum Watermarking Transform to frequency space [Cox et al. 97] Transform to frequency space [Cox et al. 97] frequency domain image DCT

11 Spread-Spectrum Image basis function DCT coefficient Salient features largest coefficients Perturb coefficients slightly to embed signal Salient features largest coefficients Perturb coefficients slightly to embed signal

12 Our Approach Extend spread-spectrum method to meshes Problem: no DCT Solution: multiresolution representation Problem: no natural sampling Solution: registration & resampling Extend spread-spectrum method to meshes Problem: no DCT Solution: multiresolution representation Problem: no natural sampling Solution: registration & resampling

13 Replacing DCT Basis Functions Multiresolution frequency information Progressive mesh [Hoppe 96] Multiresolution frequency information Progressive mesh [Hoppe 96] cosine basis image mesh ?

14 Multiresolution Neighborhoods Naturally correspond to important features Provide hints on allowable perturbation Naturally correspond to important features Provide hints on allowable perturbation corresponding mesh region vertex neighborhood

15 Scalar Basis Function i amplitude i direction d i radius displacement

16 Watermark Insertion Construct basis functions 1 … m

17 Watermark Insertion Construct basis functions 1 … m Perturb each vertex: Construct basis functions 1 … m Perturb each vertex: watermark coefficient basis function coefficient watermark direction Matrix system:

18 Watermark Extraction Get points v* on attacked mesh surface corresponding to original mesh vertices v Use same basis functions 1 … m and hence same matrix B Solve least-squares system for w*: Get points v* on attacked mesh surface corresponding to original mesh vertices v Use same basis functions 1 … m and hence same matrix B Solve least-squares system for w*:

19 False-Positive Probability Correlation = Correlation = P fp computed from and m using Students t-test Declare watermark present if P fp < P thresh ( e.g. P thresh = ) Correlation = Correlation = P fp computed from and m using Students t-test Declare watermark present if P fp < P thresh ( e.g. P thresh = )

20 Process (1) original mesh(2) watermarked(exaggerated) (3) suspect mesh(4) registered(5) resampled

21 Registration & Resampling Registration: [Chen & Medioni 92][Chen & Medioni 92] Resampling choices: Closest point projectionClosest point projection Ray-casting along local normalRay-casting along local normal Global deformation of originalGlobal deformation of originalRegistration: [Chen & Medioni 92][Chen & Medioni 92] Resampling choices: Closest point projectionClosest point projection Ray-casting along local normalRay-casting along local normal Global deformation of originalGlobal deformation of original

22 Global Deformation Deform original mesh to fit suspect mesh Minimize: +Inter-mesh distance ( vertex springs ) +Deformation ( edge springs ) +Penalty for flipped triangles Accurate, but slow Deform original mesh to fit suspect mesh Minimize: +Inter-mesh distance ( vertex springs ) +Deformation ( edge springs ) +Penalty for flipped triangles Accurate, but slow Optimized mesh Suspect mesh

23 Results similarity 1/2 faces watermarked mesh watermarked mesh 2 nd watermark noise

24 Results smoothing all attacks watermarked mesh cropped 1/8 faces

25 Summary Robust watermarking for 3D meshes Spread-spectrumSpread-spectrum Basis functions from multiresolution analysisBasis functions from multiresolution analysis Resampling as global optimizationResampling as global optimization Resilient to a variety of attacks Robust watermarking for 3D meshes Spread-spectrumSpread-spectrum Basis functions from multiresolution analysisBasis functions from multiresolution analysis Resampling as global optimizationResampling as global optimization Resilient to a variety of attacks

26 Future Work Consider other attacks: General affine and projective transformsGeneral affine and projective transforms Free-form deformations! [StirMark by Petitcolas]Free-form deformations! [StirMark by Petitcolas] Explore other basis functions e.g. [Guskov et al. 99]e.g. [Guskov et al. 99] Fast mesh recognition web crawler Consider other attacks: General affine and projective transformsGeneral affine and projective transforms Free-form deformations! [StirMark by Petitcolas]Free-form deformations! [StirMark by Petitcolas] Explore other basis functions e.g. [Guskov et al. 99]e.g. [Guskov et al. 99] Fast mesh recognition web crawler


Download ppt "Robust Mesh Watermarking Emil Praun Hugues Hoppe Adam Finkelstein Emil Praun Hugues Hoppe Adam Finkelstein Princeton University Microsoft Research Princeton."

Similar presentations


Ads by Google