Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

An Watermark Embedding System that Maximizes the Detection Strength after Post-processed

IP.com Disclosure Number: IPCOM000016503D
Original Publication Date: 2003-Jun-26
Included in the Prior Art Database: 2003-Jun-26
Document File: 2 page(s) / 16K

Publishing Venue

IBM

Abstract

Disclosed is an watermark embedding system and method that maximizes the detection strength(value) after post-processing (PP), such as compression, Digital-to-Analog / Analog-to-Digital(DA/AD) conversion, filtering, and so on, without affecting the fidelity.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 51% of the total text.

Page 1 of 2

  An Watermark Embedding System that Maximizes the Detection Strength after Post-processed

  Disclosed is an watermark embedding system and method that maximizes the detection strength(value) after post-processing (PP), such as compression, Digital-to-Analog / Analog-to-Digital(DA/AD) conversion, filtering, and so on, without affecting the fidelity. This invention has following points and advantages:

1. When PP after watermark is embedded can be predicted, the detection strength can be maximized after the PP takes place.
2. The detection strength doesn't get weaker even if the prediction doesn't hit.
3. The detection strength is improved even if the prediction of the parameter of the PP doesn't precisely hit as far as the prediction itself hits, such that prediction " compression" hits but compression ratio doesn't.
4. The fidelity, such as PSNR for still images and motion pictures, is not affected by utilizing this method. PSNR is calculated as:
PSNR= 10log10(N*Imax^2/Σ_x,y (I'(x,y)-I(x,y))^2) (dB) , (1) where N is the number of pixels, Imax is maximum luminance, usually 255, and I'(x,y) and I(x,y) are pixel values after and before watermark is embedded, respectively.
5. Multiple PPs can be supported, i.e.when more than one predicted post-processings exist, detection value can be maximized after each post-processing takes place.

Following is the detailed description of this method: Detection value of the watermark is calculated as follows: (hereafter, bold letters are for vectors)

D(X) = X・W/(¦¦X¦¦*¦¦W¦¦) , (2) where X, W are n dimensional vectors. X={X1,X2,..,Xn} is extracted from the target content, W is a predefined vector, ・stands for inner product, and * stands for normal multiplication. Each Xi is calculated as Xi = Σ_j x_ij, where x_ij (j=1,2,..,m) are sample values, such as DCT coefficients after +1 or -1, which is generated by very long pseudo random sequence, multiplied. PPs can be classified into two, predicted PP (PPP) and unpredicted PP (UPP). Assuming that X is changed to Y after embedded, and further to Z after PPPs take place, and R (approximated by random noise) is added as the UPP vector, expected detection value after embedded, PPPs and UPP take place in this order, D(Z+R), can be written as: D(Z+R) = ¦¦Z¦¦cos(τ)/sqrt(¦¦Z¦¦^2+¦¦R¦¦^2) , (3) whereτ is the angle between Z and W. Then, the problem of maximizing D(Z+R) large can end up with that of maximizing ¦¦Z¦¦ and cos(τ), respectively, regardless of R. Following is the procedure to solve this problem:
1. Grouping: in case K PPPs exist, m samples for Xi is divided in K subgroups, {Xi1,Xi2,.., XiK}.
2. Embedding formula selection (to make ¦¦Z¦¦ large): Based on the predicted post-processing, the f...