Watermark embedding and detecting methods, systems devices and components

• US 20100226527A1
• Filed: 03/05/2010
• Published: 09/09/2010
• Est. Priority Date: 08/05/2004
• Status: Active Grant

• Current Assignee: Kan Li, Xiao-Ping Zhang
• Original Assignee: Kan Li, Xiao-Ping Zhang
• Inventors: Li, Kan, Zhang, Xiao-ping
• Granted Patent: US 8,374,384 B2
• Time in Patent Office: Days
• 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

Method determining lookup table (“LUT”) for embedding watermark. For each quantization cell, calculating probabilities that signal point falls into cell. Selecting cell by probabilities. Setting LUT value to watermark value with largest probability, subject to run constraint. For remaining cells, repeating selecting and setting steps. Other method determining quantization ensemble by calculating probability density function for signal points where the watermark value to be embedded. Distortion and robustness functions are formulated. Given robustness or distortion is selected. Functions optimized, and ensemble of quantizers determined with parameters that comply. Other method quantizing in association with lossy compression. Strength of compression determined. Adapting strength of watermark with strength of compression by a mapping. Other method selecting points for embedding watermark. Determine threshold between large and small signal points using statistical method. Select signal points for embedding according to threshold. Also, processors, computer programs, and systems.

Claims

1. A method of determining a quantization ensemble of quantizers for use in quantizing a signal for watermark embedding, the method comprising:

a) for each watermark value, calculating a probability density function for signal points where the watermark value is to be embedded,

b) based on the calculated probability density functions, formulating a distortion function and a robustness function for an ensemble of quantizers,

c) selecting a given robustness or a given distortion for the distortion and robustness functions,

d) if a given robustness is selected then;

i) minimizing the distortion function for the given robustness,

ii) determining an ensemble of quantizers with parameters that comply with the minimized distortion function, the given robustness, and the robustness function,

e) and if a given distortion is selected then

i) maximizing the robustness function for the given distortion,

ii) determining an ensemble of quantizers that complies with the maximized robustness function, the given distortion, and the distortion function.

4. The method of claim 1, wherein minimizing the distortion function and maximizing the robustness function comprise:

A. for each watermark value, determining an initial quantizer for the signal points to be embedded with the watermark value,

B. formulating a Lagrangian function J for the initial quantizers, where if a given robustness is selected then J=D−

λ

Δ

R, and where if a given robustness is selected then J=−

R+Δ

D, and R is the robustness function, D is the distortion function and λ

is a constant to be determined, and

C. minimizing function J,

and wherein, determining an ensemble of quantizers that complies with the maximized robustness function, the given distortion, and the distortion function comprises;

D. determining an ensemble of quantizers that complies with the minimized J function.

5. The method of claim 4, wherein determining an initial set of quantizers comprises determining a set of minimum mean square error (“

MMSE”

) quantizers for signal points to be embedded with watermark values.

6. The method of claim 1 wherein the distortion function D is determined as the average distortion of the ensemble of quantizers.

7. The method of claim 1, wherein the distortion function D is:

$D^{\left(v\right)}=\underset{m=1}{\overset{M-1}{\sum <br><br>}}<br><br><br><br>D_m^{\left(v\right)}=\underset{m=1}{\overset{M-1}{\sum <br><br>}}<br><br><br><br>\underset{k=1}{\overset{K}{\sum <br><br>}}<br><br><br><br>{\int <br><br>}_{d_{k-1}}^{d_k}<br><br>{<br><br>s_m-r_{m,k}<br><br>}^v<br><br>f_m<br><br>\left(s_m\right)<br><br><br><br><br><br>s_m,$

where M is the number of watermark values and f_m is probability density function of signal points in signal s to be embedded with a specific watermark value m.

8. The method of claim 1, wherein the robustness function is determined as the overall mean squared distance between adjacent quantizers.

9. The method of claim 1, wherein the robustness function is

$R=\sum _{}^{}$

k = 1 K 



[ ∑

i = 0 M - 1 



∑

j = 0 , j ≠

i M - 1 





r i , k - r j , k 

2 

P i , k ]

where r_i,c is a k-th reconstruction points of a quantizer for an i-th watermark value, P_i,k is a probability that a signal (i.e., s₁) with signal points to be embedded with the i-th watermark value, falls into a k-th quantization cell of an i-th quantizer Q_i expressed as

P^i,k=∫

<sub>di,k−

1</sub>^di,kf_i(s₁)ds₁, i=0, . . . , M−

1,

where f_i is a probability density function of the signal having signal points to be embedded with the i-th watermark value.

10. The method of claim 1, wherein the watermark values are binary and the distortion function is formulated as a function of the adjustment of an initial MMSE quantizer ensemble,

$D=D_{0,\min }+\sum _{}^{}$

k = 1 K 



P 0 , k 

τ

0 , k + D 1 , min + ∑

k = 1 K 



P 1 , k 

τ

1 , k

D_0,min is a minimum distortion function of the initial quantization ensemble corresponding to a first binary watermark bit, D_1,min is a distortion function of the initial quantization ensemble corresponding to a second binary watermark bit, K is the number of quantization levels, P_0,k is a probability that a signal containing signal points to be embedded with the first binary watermark value falls into a k-th quantization cell, P_1,k is a probability that a signal containing signal points to be embedded with the second binary watermark value falls into the k-th quantization cell, τ

_0,k and τ

_1,k are differences between new reconstruction points and initial reconstruction points of two quantizers for the first and second binary watermark bits, respectively, and represent an adjustment of the initial quantization ensemble to the determined quantization ensemble.

11. The method of claim 1 wherein the watermark values are binary and the robustness function is formulated as a function of the adjustment of the initial quantization ensemble, i.e.,

$R=\sum _{}^{}$

k = 1 K 



[ P 0 , k 

( τ

0 , k + ρ

k + τ

1 , k ) 2 ] - ∑

k = 1 K 



[ P 1 , k 

( τ

0 , k + ρ

k + τ

1 , k ) 2 ]

where P_0,k is a probability that a signal containing signal points to be embedded with the first binary watermark value falls into a k-th quantization cell, P_1,k is a probability that a signal containing signal points to be embedded with the second binary watermark value falls into the k-th quantization cell, τ

_0,k and τ

_1,k are differences between new reconstruction points and initial reconstruction points of two quantizers for the first and second binary watermark bits, respectively, and represent an adjustment of the initial quantization ensemble to the determined quantization ensemble, and ρ

_k is an initial distance between the k-th reconstruction points of the initial and determined quantization ensembles for the first and second binary watermark values.

12. The method of claim 4 wherein the initial quantization ensemble is initialized as a set of minimum mean-square-error quantizers (“

MMSE”

) according to;

$d_k=\frac{r_k+r_{k+1}}{2},k=0,\mathrm{\dots <br><br>}<br><br>,K$ $r_k=\frac{{\int <br><br>}_{d_{k-1}}^{d_k}<br><br>{\mathrm{sf}}_s<br><br>\left(s\right)<br><br><br><br><br><br>s}{{\int <br><br>}_{d_{k-1}}^{d_k}<br><br>f_s<br><br>\left(s\right)<br><br><br><br><br><br>s},k=1,\mathrm{\dots <br><br>}<br><br>,K$

where s is the source signal on which the MMSE quantizers are to be determined, f_s(s) is the probability density function of s, d_k-1, d_k and r_k are decision levels and reconstruction points, k identifies a quantization cell, and K is the total number of quantization cells.

13. The method of claim 8, wherein the minimum distortion measure of the initial quantization ensemble is determined according to:

$D_0=\underset{k=1}{\overset{K}{\sum <br><br>}}<br><br><br><br>\left[{\int <br><br>}_{d_{k-1}}^{d_k}<br><br>{<br><br>s-r_k<br><br>}^2<br><br>f_s<br><br>\left(s\right)<br><br><br><br><br><br>s\right]$

where s is the source signal on which the MMSE quantizers are to be determined, f_s(s) is the probability density function of s, d_k-1, d_k and r_k are decision levels and reconstruction points, k identifies a quantization cell, and K is the total number of quantization cells.

14. A method of embedding a watermark in a signal, the method comprising:

a) embedding the watermark in the signal using a quantization ensemble determined in accordance with the method of claim 1.

15. The method of claim 14, wherein, prior to quantizing the signal, a transform is applied to the signal and quantizing and embedding are performed on the transformed signal, and, after embedding, an inverse transform is applied to the quantized embedded transformed signal.

2. A method of determining a quantization ensemble of quantizers for use in quantizing a signal for watermark embedding, the method comprising:

a) for each watermark value, calculating a probability density function for signal points where the watermark value is to be embedded,

b) based on the calculated probability density functions, formulating a distortion function and a robustness function for an ensemble of quantizers,

c) selecting a given robustness for the distortion and robustness functions,

d) minimizing the distortion function for the given robustness, and

e) determining an ensemble of quantizers with parameters that comply with the minimized distortion function, the given robustness, and the robustness function.

3. A method of determining a quantization ensemble of quantizers for use in quantizing a signal for watermark embedding, the method comprising:

a) for each watermark value, calculating a probability density function for signal points where the watermark value is to be embedded,

b) based on the calculated probability density functions, formulating a distortion function and a robustness function for an ensemble of quantizers,

c) selecting a given distortion for the distortion and robustness functions,

d) maximizing the robustness function for the given distortion,

e) determining an ensemble of quantizers that complies with the maximized robustness function, the given distortion, and the distortion function.