# 1 multimedia security part ii mesh watermarking overview 2008/07/17 presenter ： ming-zhou liao...

TRANSCRIPT

1

Multimedia Security Part IIMultimedia Security Part IIMesh Watermarking Overview Mesh Watermarking Overview

2008/07/17Presenter ： Ming-Zhou Liao

Intelligent System/Media Processing Laboratory, Department of Computer Science and Information

Engineering, National Cheng Kung University

2

OutlineOutline

------------------------------Part I-------------------------------IntroductionBasic Concepts of 3D WatermarkingTwo Classical Scheme of Robust and Fragile Watermarking------------------------------Part II------------------------------Hierarchical Blind Watermarking of 3D Triangular Meshes

Wang, Kai; Lavoue, Guillaume; Denis, Florence; Baskurt, Atilla; Multimedia and Expo, 2007 IEEE International Conference

------------------------------Part III------------------------------ConclusionRelative Reference ResourcesHomework!?

3

OutlineOutline

------------------------------Part I-------------------------------IntroductionBasic Concepts of 3D WatermarkingTwo Classical Scheme of Robust and Fragile Watermarking------------------------------Part II------------------------------Hierarchical Blind Watermarking of 3D Triangular Meshes

Wang, Kai; Lavoue, Guillaume; Denis, Florence; Baskurt, Atilla; Multimedia and Expo, 2007 IEEE International Conference

------------------------------Part III------------------------------ConclusionRelative Reference ResourcesHomework!?

4

MotivationMotivation

“Everything” is digital these daysA copy of a digital media element is identical to the originalHow can an owner protect their content?What does all of this mean in terms of law?

What does it mean to own “bits”?

The digital rights management problem (DRM) has been addressed for many information types

5

What Is Multimedia Security?What Is Multimedia Security?

What do we want from a security system?Access and copy control

Playback controlRecord controlGeneration control...

6

What Is Multimedia Security? (Cont.)What Is Multimedia Security? (Cont.)

Goals of cryptography for classical data streams:ConfidentialitySignature & watermarkAuthentication Identifications and access control Non-repudiations

7

What Is Multimedia Security? (Cont.)What Is Multimedia Security? (Cont.)

Goals of cryptography for multimedia streams:Multimedia confidentiality (Encryption)Multimedia signature & watermarkMultimedia authentication Multimedia identifications and access control Multimedia non-repudiations

8

What Is Multimedia Security? (Cont.)What Is Multimedia Security? (Cont.)

UnfortunatelyProblems of multimedia security are the same problem of cryptography…Will it be illegal to remove security features from a data element?

In other words, security of multimedia systems depends on security of cryptographic systems

9

Typical Cryptography SystemTypical Cryptography SystemTrusted UsersTrusted Users

Figure 1. Typical cryptography system – trusted users

10

Typical Cryptography SystemTypical Cryptography System Distrusted Users Distrusted Users

Figure 2. Typical cryptography system – distrusted users

11

Media Elements Media Elements

Covered from many different fields:AudioVideoDocuments (including HTML documents)ImagesPrograms (executable code)Graphics

12

Multimedia Security - Tools Set Multimedia Security - Tools Set

There are some ways which can achieve the multimedia security:

EncryptionAuthenticationHashingTime-stampingWatermarking

13

Multimedia Security - Applications Multimedia Security - Applications

Some associated applications are addressed as follow:

PrivacyForgery detection (watermarking)Copyright protection (watermarking)Proof of purchaseProof of deliveryIntruder detection

14

What Is Watermarking? What Is Watermarking?

The use of perceptually invisible authentication techniques

“Controlled” distortion is introduced in a multimedia elementVisible watermarks also exists

GoalsVerify the owner of a digital dataDetect forgeries of an original dataIdentify illegal copies of the dataPrevent unauthorized distribution

15

What Is Watermarking? (Cont.)What Is Watermarking? (Cont.)

Main principlesTransparency

The watermark is not visible in the image under typical viewing conditions

Robustness to attackThe watermark can still be detected after the image has undergone linear and/or nonlinear operationsRobustness is crucial to the success of data embedding

CapacityThe technique is capable of allowing multiple watermarks to be inserted into the image with each watermark being independently verifiableShould be embed a nontrivial amount of information into models

16

Two Type of WatermarkingTwo Type of Watermarking

Robust watermarkingDesigned for the sender to check the copyright ownership

Fragile watermarkingDesigned for the receiver to verify the authentication of the received data Major functions

Integrity checkingChanged region locating

17

Two Type of Watermarking (Cont.)Two Type of Watermarking (Cont.)

Table 1 A comparison of fragile watermarking and digital signature applied in 3D model.

18

Extraction StrategiesExtraction Strategies

Private watermarkingNeeds original model and watermarks

Public (blind) watermarkingCan extract watermarks in the absence of the original model and the watermarksAll fragile watermarking are public watermarking scheme

Semi-public watermarkingDoes not need original model but the original watermarks are necessary for comparing stage

19

OutlineOutline

------------------------------Part I-------------------------------IntroductionBasic Concepts of 3D WatermarkingTwo Classical Scheme of Robust and Fragile Watermarking------------------------------Part II------------------------------Hierarchical Blind Watermarking of 3D Triangular Meshes

Wang, Kai; Lavoue, Guillaume; Denis, Florence; Baskurt, Atilla; Multimedia and Expo, 2007 IEEE International Conference

------------------------------Part III------------------------------ConclusionRelative Reference ResourcesHomework!?

20

Embedding TargetEmbedding Target

Representation of 3D modelsPolygonal meshesCurved surfacesSolid boundVoxel enumeration

Embedding candidateGeometry is the best candidate for data embeddingNongeometrical components have higher chances of alteration or removal than geometry

21

Embedding Primitives of Polygonal ModelEmbedding Primitives of Polygonal Model

Geometrical embedding primitivesThe coordinates of points and vertices can be modified to embed dataAdvantageous to employ primitives that are Invariant to certain class of geometrical transformations

Invariant to translation and rotationLength of lineArea of a polygonVolume of a polyhedron

Invariant to rotation, uniform-scaling, and translationRatio of the areas of two polygonal

22

Embedding Primitives of Polygonal Model Embedding Primitives of Polygonal Model (cont.)(cont.)

Topological embedding primitivesWatermark can be embedded by changing the topology of a modelThe changes may also involve change in geometry as a side effect

E.g. inserting or displacing vertices

Figure 3. Simple examples of topological embedding primitives.

23

Embedding Primitive ArrangementEmbedding Primitive Arrangement

Image or audio data already have regular implicit ordering of embedding primitives

Topological arrangementEmploys topological adjacency to arrange embedding primitivesNot resistant to topological modification

Quantitative arrangementEmploys inequality relations among the quantities

E.g. Area of triangles, volumes of polyhedrons

24

Common Attack Strategies Common Attack Strategies

The watermarking approach should withstand all of possible alterations ideally

Similarity transformsRotationUniform scaling Translation

Affine transforms Non-uniform scalingProjective distortions

Noising and de-noisingWhite Gaussian noise additionLaplacian smoothing

Figure 4. Adding random noise.

25

Common Attack Strategies (Cont.)Common Attack Strategies (Cont.)

Connectivity attacksVertex reorderingRe-triangulation

Re-samplingSimplificationMesh refinementRemeshing (introduced later)

Topological attacksCropping

Compression attacksPoint coordinates quantizationFormat conversion attacks

Figure 5. Simplification of triangular meshes.

Figure 6. Enhancement of the bunny.

26

Common Attack Strategies (Cont.)Common Attack Strategies (Cont.)

Geometrical deformationsBendingMesh editingMesh morphingLocal deformations

Figure 7. Watermarked

models and various attacks.

27

RemeshingRemeshing

The process refers to approximation such geometry using a mesh with (semi)-regular connectivity

Figure 8. A series of surface covered with regular or semi-regular regions.

28

Regularity of MeshesRegularity of Meshes

In topological mesh modeling our only concern is mesh structure

How faces, edges and vertices are related with each other

Topological regularity as all faces and vertices have the same combinatorial property

i.e. having the same size and same valence, respectivelyThe valence of a vertex is defined as the number of edge-ends that emanates from that vertexThe face size is counted as the number of edge sides (also called half-edges) belonging to a face

29

Regularity of Meshes (Cont.)Regularity of Meshes (Cont.)

Figure 9. A series of examples of regular mesh object.

30

OutlineOutline

31

Triangle Similarity Quadruple AlgorithmTriangle Similarity Quadruple Algorithm

Pair of dimensionless quantities{ b/a, h/c } or { θ1, θ2 }

Micro-Embedding Primitive (MEP)Each MEP stores a quadruple of values {Maker, Subscript, Data1, Data2}

Figure 11. A macro-embedding-primitive.

Figure 10. Examples of dimensionless quantities that define a set of similar triangles.

32

Classical Strategy of Fragile Watermarking Classical Strategy of Fragile Watermarking

Computes two indices for every vertexLocation indexValue indexHash function

Perturbed every vertex to make two indices equalFigure 12. An example of triangular mesh topology.

Figure 13. Modify the coordinate of vertex to match the value of hash function.

33

Problems of These ApproachProblems of These Approach

TSQ Algorithm Resistant

RotationTranslationUniform scaling

Classical fragile watermarking Causality problem

Former processed vertex will be changed by later processed neighboring vertex

Convergence problemUser can not control the distortion induced by perturbing process

34

OutlineOutline

35

Introduction of The PaperIntroduction of The Paper

Embed two watermark into different resolution levelsGeometrically robust watermark

Synchronizing and quantizing watermark primitives according to the edge length of the coarsest level

High-capacity watermarkPermutation of the norms of a group of wavelet coefficients

Both watermark are blind and invariant to similarity transformations

36

Proposed FrameworkProposed Framework

Pre-procedureThe irregular mesh is remeshed to generate a corresponding semi-regular mesh

Normally has a negligible geometrical distortion More favorable for compression thanks to its simple connectivity

Extraction Suppose that a mesh with the same semi-regular connectivity can be reconstructed

37

Proposed Framework (Cont.)Proposed Framework (Cont.)

Lazy wavelet decomposition mechanismA group of four triangles is merged in oneThree of the six initial vertices are conserved

in the lower resolution

This kind of wavelet analysis is applicable only on semi-regular triangular meshes

Figure 14. Illustration of 3D triangular mesh wavelet analysis.

38

Proposed Framework (Cont.)Proposed Framework (Cont.)

Multiresolution analysis is a very suitable tool for hierarchical watermarking

No inter-infection between different watermarks if they are inserted in different levels

Figure 15. Hierarchical watermarking of a semi-regular mesh.

39

Robust and Blind WatermarkingRobust and Blind Watermarking

All the norms are altered to locate in the middle of the nearest correct subinterval

The synchronizing mechanism and the watermarking primitives are separated

The causality problem is avoided

Figure 16. Three quantization examples of coefficient norms.

40

High-Capacity Blind WatermarkingHigh-Capacity Blind Watermarking

High-capacity watermark embedding mechanismThe wavelet coefficients are indexed according to the lengths of their associated edgesThe order sequence of coefficients is a permutation of the n numbers thus has n! possibilitiesSort and modify the residues of the norms to achieve a better imperceptibility

41

High-Capacity Blind Watermarking (Cont.)High-Capacity Blind Watermarking (Cont.)

The modified norm is determined by

Table 2. Example of high-capacity watermark embedding steps.

42

High-Capacity Blind Watermarking (Cont.)High-Capacity Blind Watermarking (Cont.)

The edges can be divided into several groupsMake the watermark less fragileAvoid the possible floating number calculation errors

Correspondence between bit strings and possible permutation has to be established

For two permutationsThe one with a bigger first number represents a bigger bit stringIf the first number is the same then compare the second number, and so on

1, 2, 3, …, n-1, n 0, 0, …, 0, 01, 2, 3, …, n, n-1 0, 0, …, 0, 1

43

Experimental ResultsExperimental Results

Figure 17. Capacity comparison of different methods.

44

Experimental Results (Cont.)Experimental Results (Cont.)

Figure 18. Hierarchical watermarking example: (a) original semi-regular mesh (b) watermarked mesh

Figure 19. Four attacked watermarked meshes: (a) 0.4% additive noise (b) 6 average smoothing (c) enhancement (d) 7-bit quantization

45

Experimental Results (Cont.)Experimental Results (Cont.)

Table 3. Results of robustness experiments

46

Comments Comments

Need to prepare the remeshing techniques that are insensitive to connectivity changes

The high-capacity watermark is somewhat fragile

The “robust” watermark can’t withstand the local deformation and cropping

47

OutlineOutline

48

Conclusion Conclusion

Watermarking techniques for 3D model face many challenges for the follow reasons

Compared with images, only a small amount of data is available for watermark embeddingNo unique representation nor implicit ordering of model data existsNo robust transformation field could be used to embed watermark

Multiresolution representation may overcome this issue

49

Conclusion (Cont.) Conclusion (Cont.)

Watermarking is still an interesting research area with many interesting problems:

Where will it be useful?Will watermarking only be used a second-tier security systemWill there be significant theoretical developments?

50

OutlineOutline

51

Relative Reference ResourcesRelative Reference Resources

JournalComputerized Medical imaging and GraphicsCVGIP: Computer Vision and Image UnderstandingCVGIP: Graphical Models and Image ProcessingDigital Signal Processing: A Review JournalIEE Proceedings Part I: Communications Speech and VisionIEEE Trans. Image ProcessingIEEE Trans. Pattern Analysis and Machine IntelligenceIEEE Trans. Systems, Man, and CyberneticsImage and Vision ComputingInt’I Journal of Computer Vision

52

Relative Reference Resources (Cont.)Relative Reference Resources (Cont.)

Int’I Journal on Document Analysis and RecognitionInt’I Journal of Pattern Recognition and Artificial IntelligenceJournal of Electronic ImagingJournal of Imaging Science and TechnologyJournal of Imaging TechnologyJournal of Mathematical Imaging and VisionJournal of the Optical Society of America A: Optics Image ScienceJournal of Vision and ApplicationsMultidimensional Systems and Signal Processing

53

Relative Reference Resources (Cont.)Relative Reference Resources (Cont.)

Optical EngineeringPattern Analysis and ApplicationsPattern RecognitionPattern Recognition LettersPattern Recognition and Image Analysis (Moscow)Signal Processing (EURASIP)Signal Processing: Image Communication (EURASIP)The Transactions of The IEICE Series (Japan)

54

Relative Reference Resources (Cont.)Relative Reference Resources (Cont.)

Proc. of conferences(3DIM) 2th Int’I Conf. on 3D Digital Imaging and Modeling(ACCV) Asian conf. Computer Vision(ASSP) IEEE Conf. Acoustics, Speech and Signal Processing(CVPR) IEEE Conf. Computer Vision and Pattern Recognition(EMM) 2th Int’I WK. on Energy Minimization Methods in CVPR(ICAIP) 8th Int’I Conf. Computer Analysis of Images and Patterns(ICCV) International Conf. Computer Vision(ICDAR) IAPR Conf. Document Analysis and Recognition

55

Relative Reference Resources (Cont.)Relative Reference Resources (Cont.)

(ICSPAT) Int’I Conf. Signal Processing Applications & Technology(IGARSS) IEEE conf. Geographic and Remote-sensing Sciences(IPA) Int’I Conf. Image Processing and its Applications(IPTA) Int’I Conf. Image Processing: Theory and Applications(VCIP) SPIE’s Symp. Visual commun. And Image Processing

56

Relative Reference Resources (Cont.)Relative Reference Resources (Cont.)

AssociationIEEE (the Institute of Electrical and Electronics Engineers)ACM (Association for Computing Machinery)IAPR (The International Association for Pattern RecognitionIS&T (The Society for Imaging Science and Technology)SPIE (The International Society for Optical Engineering)IASTED (The International Association of Science and Technology for Development)

57

OutlineOutline

58

Homework!?Homework!?

Defense groupTo do

Design a suitable image watermarking algorithm to resist other malicious manipulation

Must satisfy the follow constrains

ConstrainsPeak Signal to Noise Ratio (PSNR): 38 ↑Watermark correlation: 0.7 ↑

59

Homework!? (Cont.)Homework!? (Cont.)

Attack groupTo do

Attack the target watermarked image for disturbing watermark extraction

Must satisfy the follow constrains

ConstrainsPeak Signal to Noise Ratio (PSNR ): 38 ↑Watermark correlation: 0.7 ↓

60

Homework!? (Cont.)Homework!? (Cont.)

Recommend toolMATLAB

ISMP FTPAddress: 140.116.247.34Port: 400Account: ismpPassword: ismp 13

Target image & watermark informationDecided by defense team

61

RecruitRecruit

Benefits about being activity managerNo need to sweep the floor on the ground every weekNo need to charge the function of fluorescent light in lab

No need to borrow the key for seminar every weekNo need to get up early for setting up the stuff for seminar every weekMay be pay raise someday!?You are the only one……It’s cool………

62

Q&AQ&A

Thank you for your listening!