SECURE SPREAD SPECTRUM WATERMARKING FOR MULTIMEDIA

1

• SECURE SPREAD SPECTRUM WATERMARKING FOR
  MULTIMEDIA
• Cox, Kilian, Leighton, and Shamoon
• IEEE Transactions on Image Processing
• Vol. 6. No.12, December 1997

2
WATERMARK CHARACTERISTICS

• To be effective, a watermark should have the
  following characteristics
• Unobtrusive The watermark should be
  perceptually invisible or its presence should not
  interfere with the work being protected.
• Robustness The watermark must be difficult
  (hopefully impossible) to remove. In particular
  the watermark should be robust to the below
  attacks
• Common signal processing The watermark should
  still be retrievable even if common signal
  processing operations are applied to the data
  These include digital-to-analog and
  analog-to-digital conversion, resampling,
  requantization (including dithering and
  recompression), and common signal enhancements to
  image contrast and color.
• Common geometric distortions Watermarks in
  image and video data should also be immune from
  geometric image operations such as rotation,
  translation, cropping, and scaling.
• Subterfuge Attacks (Collusion and Forgery)
• In addition the watermark should be robust to
  collusion by multiple individuals who each
  possess a watermarked copy of the data That is
  the watermark should be robust to combining
  copies of the same data set to destroy the
  watermarks.
• Further if a digital watermark is to be used in
  litigation, it must be impossible for colluders
  to combine their images to generate a dierent
  valid watermark with the intention of framing a
  third party.

3
WATERMARK CHARACTERISTICS

• Universality The same digital watermarking
  algorithm should apply to all three media under
  consideration. This is potentially helpful in
  the watermarking of multimedia products. Also
  this feature is conducive to implementation of
  audio and image/video watermarking algorithms on
  common hardware.
• Unambiguousness Retrieval of the watermark
  should unambiguously identify the owner.
  Furthermore, the accuracy of owner identification
  should degrade gracefully in the face of attack.
• There are two parts for building a strong
  watermark
• Watermark structure
• Insertion strategy
• Two arguments to make the watermark robust and
  secure
• The watermark must be placed explicitly in the
  perceptually most significant components.
• The watermark must be composed of random numbers
  drawn from a Gaussian distribution (N(0,1), where
  N(µ,s2) denotes a normal distribution with mean
  µ, and variance s2).

These two components must be designed correctly.
4
WATERMARK EMBEDDING AND DETECTION

• Watermark embedding
• Compute the NxN DCT of an NxN gray scale cover
  image I.
• Embed a sequence of real values X x1,x2,, xn,
  according to N(0,1), into the n largest magnitude
  DCT coefficients, excluding the DC component
• Compute the inverse DCT to obtain the watermarked
  cover image I.
• Watermark detection
• Compute the DCT of the watermarked (and possibly
  attacked) cover image I.
• Extract the watermark X
• Evaluate the similarity of X and X using
• sim (X,X)
• If sim (X,X) gt T, a given threshold, the
  watermark exists.

5
EXPERIMENTS

• The Discrete Cosine Transform (DCT)
• The DCT represents an image as a sum of sinusoids
  of varying magnitudes and frequencies.
• The DCT has the property that, for a typical
  image, most of the visually significant
  information about the image is concentrated in
  just a few coefficients of the DCT.
• The DCT is at the heart of the international
  standard lossy image compression algorithm known
  as JPEG.
• In all experiments, a watermark length of 1000
  was used.
• The watermark was added to the image by
  modifying 1000 of the more perceptually
  significant DCT coefficients of the image.
• The DC term was excluded in the embedding
  process.
• The value of the scaling factor a was 0.1.

6
ATTACKS
7
EMBEDDING THE WATERMARK
8
DETECTING THE WATERMARK
9
ORIGINAL AND WATERMARKED IMAGES
10
EXPERIMENTS - Uniqueness of watermark
32
The response of the watermark detector for the
real watermark is 32.0.
11
EXPERIMENTS Image scaling
The response of the watermark detector for the
real watermark is 13.4. The response is higher
than those of the fake watermarks!
12
EXPERIMENTS JPEG compression
The response of the watermark detector for the
real watermark is 22.8. The response is higher
than those of the fake watermarks!
The response of the watermark detector for the
real watermark is 13.9. The response is higher
than those of the fake watermarks!
13
EXPERIMENTS Dithering
The response of the watermark detector for the
real watermark is 5.2. The response is higher
than those of the fake watermarks!
14
EXPERIMENTS Cropping
The response of the watermark detector for the
real watermark is 14.6. The response is higher
than those of the fake watermarks!
15
EXPERIMENTS Cropping
The response of the watermark detector for the
real watermark is 10.6. The response is higher
than those of the fake watermarks!
16
EXPERIMENTS Print, xerox, and scan
The response of the watermark detector for the
real watermark is 4.0. The response is still
higher than those of the fake watermarks!
17
EXPERIMENTS Rewatermarking
18
EXPERIMENTS Rewatermarking
19
EXPERIMENTS Collusion
20
EXPERIMENTS Collusion
21
CONCLUSIONS

• The watermark is consisted of a k random numbers
  drawn from a N(0,1) distribution.
• The watermark is embedded in the perceptually
  most significant components.
• This maximizes the change of detecting the
  watermark after normal A/V processes and
  intentional attacks.
• In the experiments, the watermark was added to
  the largest 1000 DCT coefficients.
• The cover image was the Bavarian couple.
• Attacks
• Scaling
• JPEG compression
• Dithering
• Cropping
• Printing, xeroxing, and scanning
• Rewatermarking
• Collusion
• After all attacks, the correlation with the real
  watermark has a peak.