Digital watermarking in a perceptually uniform domain

• US 6,792,129 B1
• Filed: 08/23/2000
• Issued: 09/14/2004
• Est. Priority Date: 08/25/1999
• Status: Expired due to Term

• Current Assignee: Rakuten Group, Inc.
• Original Assignee: Sharp Laboratories of America Incorporated (Hon Hai Precision Industry Co., Ltd.)
• Inventors: Lei, Shawmin, Zeng, Wenjun
• Primary Examiner(s): Au, Amelia M.
• Assistant Examiner(s): KIM, CHONG R
• Application Number:
  US09/645,105
• Time in Patent Office: 1,483 Days
• Field of Search: 382/100, 382/248, 382/250, 382/274
• US Class Current: 382/100
• CPC Class Codes: G06T 1/0028 Adaptive watermarking, e.g. Human Visual System [HVS]-based watermarking
• CPC Class Codes: G06T 2201/0052 Embedding of the watermark in the frequency domain
• CPC Class Codes: G06T 2201/0202 whereby the quality of watermarked images is measured; Measuring quality or performance of watermarking methods; Balancing between quality and robustness

Abstract

Methods and apparatus for watermarking a digital media object, and for detecting watermarks, are presented. The basic concept underlying the disclosed approach is watermarking/detection in a transform space that allows the same level of watermarking to be applied to all samples. For instance, in one embodiment, a watermarking system first nonlinearly transforms the original signal to a perceptually uniform domain, and then embeds the watermark in this domain without varying the statistical properties of the watermark at each sample. At the watermark detector, a candidate image is transformed to the same perceptually uniform domain, and then correlated with the original watermark sequence. Under such conditions, it is shown that an optimal watermark detector can be derived. This approach is particularly attractive when the original image is unavailable at the detector, as it effectively prevents the image content from biasing the watermark detection score.

Claims

1. A method of detecting a watermark in a digital media object comprising:

for a set of features {x_i*} extracted from a digital data set, transforming each feature x_i* to a corresponding perceptual domain feature z_i* in a perceptual domain feature set {z_i*};

providing a set of pseudorandom numbers {S_1i} derived from a selected watermark key; and

calculating a correlation figure q for {z_i*} and {S_1i};

wherein the transforming operates on a set of frequency-based coefficients, transforming a frequency-based coefficient to a corresponding perceptual domain coefficient according to at least one of a frequency sensitivity function value, a luminance sensitivity adjustment, and a self-contrast masking coefficient.

2. The method of claim 1, wherein $q=\frac{M_y}{}$

nVy,

where n is the number of features in the set {x_i*}, Y_i=z_i*S_1i, M_y is the sample mean of {Y_i}, and V_y² is the sample variance of {Y_i}.

3. The method of claim 1, wherein the digital data set comprises coefficients from a frequency-based representation of a digital media object.

4. The method of claim 3, wherein the digital media object is selected from the group consisting of a digital image, one or more frames of a digital video image sequence, digitized audio, an arrangement of graphical objects, and combinations thereof.

5. The method of claim 3, further comprising extracting a subset of the coefficients from the frequency-based representation of the digital media object as the digital data set.

6. The method of claim 5, wherein extracting a subset comprises excluding lowest-frequency coefficients from selection to the digital data set.

7. The method of claim 5, wherein extracting a subset comprises excluding highest-frequency coefficients from selection to the digital data set.

8. The method of claim 5, wherein extracting a subset comprises excluding a coefficient from selection to the subset when that coefficient has a magnitude less than a corresponding contrast masking threshold.

9. The method of claim 3, wherein the digital media object is a visual object, and wherein transforming each feature x_i* to a corresponding perceptual domain feature z_i* comprises calculating

10. The method of claim 9, wherein transforming each x_i* to a corresponding perceptual domain feature z_i* further comprises calculating

11. The method of claim 10, wherein $w_i=b$

(1+a

∑

k

 



near

 



i



xk*

β

/

Φ

i

),

where a and b are normalization factors, |Φ

_i| is the size of the neighborhood, and β

is a positive value.

12. The method of claim 3, further comprising using an original version of the digital media object during watermark detection, by providing a set of perceptual domain features {z_i} corresponding to the original version of the digital media object, and subtracting z_i from z_i* prior to calculating a correlation figure q.

13. A watermark detector comprising:

a perceptual transform to calculate a perceptual domain data set corresponding to an input digital media object; and

a correlator to calculate a correlation figure for the perceptual domain data set and a watermark signature, wherein the transform operates on a set of frequency-based coefficients, transforming a frequency-based coefficient to a corresponding perceptual domain coefficient according to at least one of a frequency sensitivity function value, a luminance sensitivity adjustment, and a self-contrast masking coefficient.

14. The watermark detector of claim 13, further comprising a pseudorandom sequence generator capable of accepting a watermark key and generating the watermark signature corresponding to that key.

15. The watermark detector of claim 14, further comprising a sequence modulator to modulate the watermark signature according to preselected signature data.

16. The watermark detector of claim 13, further comprising a frequency-based transform, wherein the input to the frequency-based transform comprises the input digital media object and the output of the frequency-based transform comprises the input to the perceptual transform.

17. The watermark detector of claim 13, wherein the perceptual transform operates on a set of frequency-based coefficients, the perceptual transform comprising a feature extractor to extract a subset of the frequency-based transform coefficients for transformation to the perceptual domain data set.

18. The watermark detector of claim 17, wherein the feature extractor excludes lowest-frequency coefficients and highest-frequency coefficients from selection to the subset.

19. The watermark detector of claim 17, wherein the feature extractor excludes a frequency-based coefficient from selection to the subset when that coefficient has a magnitude less than a corresponding contrast masking threshold.

20. The watermark detector of claim 13, wherein the perceptual transform operates on a set of frequency-based coefficients, the transform comprising a frequency sensitivity adjuster that weights a frequency-based coefficient according to a frequency sensitivity function value corresponding to the frequency represented by that coefficient.

21. The watermark detector of claim 13, wherein the perceptual transform operates on a set of frequency-based coefficients, the transform comprising a luminance sensitivity adjuster that weights a frequency-based coefficient according to a local average luminance figure.

22. The watermark detector of claim 13, wherein the perceptual transform operates on a set of frequency-based coefficients, the transform comprising a self-contrast masking adjuster that scales a frequency-based coefficient according to the ability of that coefficient to mask a signal at approximately the same frequency, orientation, and spatial location represented by that coefficient.

23. The watermark detector of claim 13, wherein the perceptual transform operates on a set of frequency-based coefficients, transforming a frequency-based coefficient x_i to a corresponding perceptual domain coefficient z_i according to the equation

24. The watermark detector of claim 23, wherein the perceptual transform further comprises a neighborhood masking adjuster that adjusts a frequency-based coefficient based on the amplitudes of surrounding coefficients.

25. A watermark detector comprising:

means for transforming a target data set into a perceptual domain data set;

means for deriving a watermark signature from a selected watermark key; and

means for testing the perceptual domain data set for the existence of the watermark signature, wherein the transforming operates on a set of frequency-based coefficients, transforming a frequency-based coefficient to a corresponding perceptual domain coefficient according to at least one of a frequency sensitivity function value, a luminance sensitivity adjustment, and a self-contrast masking coefficient.

26. An apparatus comprising a computer-readable medium containing computer instructions that, when executed, cause a processor or multiple communicating processors to perform a method for detecting a watermark in a digital media object, the method comprising:

for a set of features {x_i′

} extracted from a digital data set, transforming each feature x_i′

to a corresponding perceptual domain feature z_i′

in a perceptual domain feature set {z_i′

};

providing a set of pseudorandom numbers {S_1i} derived from a selected watermark key; and

calculating a correlation figure q for {z_i′

} and {S_1i}, wherein the transforming operates on a set of frequency-based coefficients, transforming a frequency-based coefficient to a corresponding perceptual domain coefficient according to at least one of a frequency sensitivity function value, a luminance sensitivity adjustment, and a self-contrast masking coefficient.

27. The apparatus of claim 26, wherein $q=\frac{M_y}{}$

nVy,

where n is the number of features in the set {x_i′

}, Y_i=z_i′

S_1i, M_y is the sample mean of {Y_i}, and V_y² is the sample variance of {Y_i}.

28. The apparatus of claim 26, wherein the digital data set comprises coefficients from a frequency-based representation of a digital media object.

29. The apparatus of claim 28, wherein the method further comprises extracting a subset of the coefficients from the frequency-based representation of the digital media object as the digital data set.

30. The apparatus of claim 28, wherein the digital media object is a visual object, and wherein transforming each feature x_i′

to a corresponding perceptual domain feature z_i′

comprises calculating

31. The apparatus of claim 30, wherein transforming each x_i′

to a corresponding perceptual domain feature z_i′

further comprises calculating

32. The apparatus of claim 31, wherein $w_i=b$

(1+a

∑

k

 



near

 



i



xk*

β

/

Φ

i

),

where a and b are normalization factors, |Φ

_i| is he size of the neighborhood, and β

is a positive value.