T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A November 2005Analysis of Attacks on Common Watermarking Techniques 1 A study on the robustness.

November 2005 Analysis of Attacks on Common Watermarking Techniques1

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Analysis of Attacks on Common Watermarking Techniques

A study on the robustness of typical watermarks

Johnson Lee

Prepared for EECE 575 – Digital Image Processing

November 2005 Analysis of Attacks on Common Watermarking Techniques2

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Outline

• Background Info

• Techniques– Spatial Domain– DCT Domain– Wavelet Domain

• Conclusion

November 2005 Analysis of Attacks on Common Watermarking Techniques3

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Background – Terminology

• Watermarkingtechniques that allow secret communication, usually by embedding or hiding the secret information

• Fingerprintingspecial application of watermarking related to copyright protection

• Copy Protection vs Copyright Protectioncopy protection inhibits the copy process itself while copyright protection embeds copyright information into an object

November 2005 Analysis of Attacks on Common Watermarking Techniques4

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Principles of Watermarking

• Capacity– As much information as possible

• Security– Only be accessible by authorized parties

• Robustness– Resist against attacks

• Imperceptibility– Invisible

November 2005 Analysis of Attacks on Common Watermarking Techniques5

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Principles of Watermarking

• Capacity– As much information as possible

• Security– Only be accessible by authorized parties

• Robustness Focus of this study– Resist against attacks

• Imperceptibility– Invisible

November 2005 Analysis of Attacks on Common Watermarking Techniques6

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Principles of Watermarking

• Capacity– As much information as possible

• Security– Only be accessible by authorized parties

• Robustness Focus of this study– Resist against attacks

• Imperceptibility Lightly present– Invisible

November 2005 Analysis of Attacks on Common Watermarking Techniques7

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Type of Attacks

• Unintentional– Image compression

Ideal watermarking and compression systems inherently mutually exclusive

• Malicious– Disabling watermark

Geometric distortion or noise

November 2005 Analysis of Attacks on Common Watermarking Techniques8

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Basic Concept of Watermarking

• Let O = original imageW = watermarkM = watermarked imagef(x) = some transformation

• Then,

M = O + f(W)

November 2005 Analysis of Attacks on Common Watermarking Techniques9

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

CDMA Spatial Domain Spread• Limited capacity

Ideal for small watermarks PSNR = 335.3749

November 2005 Analysis of Attacks on Common Watermarking Techniques10

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

CDMA Spatial Resistance

• JPEG resistant until quality factor 50

• No resistance against resizing, rotation

• Mildly tolerable of mean and median filters – up to [3 3] Recovered Watermark

[3 3]

Recovered Watermark

[2 2]

Common to Many

Techniques

November 2005 Analysis of Attacks on Common Watermarking Techniques11

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Bruyndonckx (Spatial Domain)• Signature generated with input text

• Based on zone/category classification to embed information Algorithm use different edge gradients to embed info

November 2005 Analysis of Attacks on Common Watermarking Techniques12

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Bruyndonckx (Spatial Domain)• Pixel classification, Zone based

PSNR = 26407` Original

November 2005 Analysis of Attacks on Common Watermarking Techniques13

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Bruyndonckx (Spatial Domain)

• JPEG resistant until quality factor 50

• No resistance against resizing, rotation

• Mildly tolerable of median and mean filters

bit matches: 491/584

correlation: 0.681507

bit matches: 515/584

correlation: 0.763699

bit matches: 421/584

correlation: 0.441781

bit matches: 439/584

correlation: 0.503425

Mean Median

[3 3]

[4 4]

50 60bit matches: 394/584 bit matches: 538/584correlation: 0.349315 correlation: 0.842466SAME PROBLEM AS

CDMA SPATIAL DOMAIN

November 2005 Analysis of Attacks on Common Watermarking Techniques14

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Comparison-based Correlation DCT Midband

November 2005 Analysis of Attacks on Common Watermarking Techniques15

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Comparison-based Correlation DCT Midband - Resistance

• JPEG resistant until quality factor 28

• Geometric – cannot be done because algorithm needs exact size

• Mean and Median – [4 4] tolerant

COMPARISON

CDMA &Bruyndonckx

JPG – 50Geo – 0

Median – [3 3]Mean – [3 3]

Mean Median

[4 4] [4 4]

November 2005 Analysis of Attacks on Common Watermarking Techniques16

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Cox (DCT Domain)

PSNR = 665.9571 Original

November 2005 Analysis of Attacks on Common Watermarking Techniques17

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Cox Resistance (DCT Domain)• Extremely robust to JPG, Q – 20

• Geometric

– cannot be done because algorithm needs exact same size as original

• Mean and Median– Highly resistance since not dependent on high freq noise

Correlation 0.999073

0.947364 0.947354

0.867709 0.949934[10 10]

[6 6]

Mean Median

November 2005 Analysis of Attacks on Common Watermarking Techniques18

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Future/In Progress Work

• Resizing needs to be redone– Most algorithms require input file to be

same size as original, can be done by resizing

• DWT Domain CDMA and advanced DWT techniques