Blind wavelet-based watermarking method
AssigneeNational Taiwan University of Science and Technology
InventorsJing-Ming Guo, Yu-Quan Tzeng
Life of this patent
Estimating the term…
Abstract
A blind wavelet-based watermarking method is provided to extract one or more embedded watermarks form one or more high subbands of a watermarked image generated by 1-level or 2 level wavelet transform. One or more least-mean-square (LMS) filters are trained to predict the data sets in the high subbands of an original image by converting a low subband of the watermarked image. Therefore, the one or more embedded watermarks could be extracted by comparing the predicted data sets in the high subbands with data sets in corresponding subbands of the watermarked image.
Claims10
A blind wavelet-based watermarking method for extracting at least one hidden watermark from a watermarked image, the method comprising:
- performing a wavelet transform to decomposed the watermarked image into a first data set in a subband LL1, a second data set in a subband LH1, a third data set in a subband HL1 and a fourth data set in a subband HH1;
- converting the first data set in the subband LL1 into a first predicted data set in the subband LH1, into a second predicted data set in the subband HL1, and into a third predicted data set in the subband HH1 via three LMS (least-mean-square) filters;
- and
- extracting the at least one hidden watermark by comparing the first predicted data set with the second data set, comparing the second predicted data set with the third data set, or comparing the third predicted data set with the fourth data set.
A blind wavelet-based watermarking method for extracting at least one hidden watermark from a watermarked image, the method comprising:
- performing a 2-level wavelet transform to decomposed the watermarked image into a first data set in a subband LL2, a second data set in a subband LH2, a third data set in a subband HL2, a fourth data set in a subband HH2, a fifth data set in a subband LH1, a sixth data set in a subband HL1, a seventh data set in a subband HH1;
- converting the first data set in the subband LL2 into a first predicted data set in the subband LH2, into a second predicted data set in the subband HL2, and into a third predicted data set in the subband HH2 via three LMS filters;
- performing an inverse wavelet transform to transform the first data set, the first predicted data set, the second predicted data set and the third predicted data set into a fourth predicted data set in the subband LL1;
- converting the fourth predicted data set in the subband LL1 into a fifth predicted data set in the subband LH1, into a sixth predicted data set in the subband HL1, and into a seventh predicted data set in the subband HH1 via other three LMS filters;
- and
- extracting the at least one hidden watermark by comparing the first predicted data set with the second data set, comparing the second predicted data set with the third data set, comparing the third predicted data set with the fourth data set, comparing the fifth predicted data set with the fifth data set, comparing the sixth predicted data set with the sixth data set, or comparing the seventh predicted data set with the seventh data set.
A method for optimizing coefficients of a least-mean-square (LMS) filter for a wavelet transform in a high subband {λ, θ}, parameter λ indicating a level of the wavelet transform, and parameter θ indicating an orientation of the wavelet transform, the LMS filter being adapted to convert a low-frequency image in a low subband into a high-frequency image in the high subband {λ, θ}, the method comprising following steps:
- (a) obtaining a reconstructed coefficient {circumflex over (d)}λ,i,jθ,k of the high subband {λ, 0} at
- ∑
- m
- ,
- n
- ∈
- R
-
- ∑
- w
- λ
- ,
- m
- ,
- n
- θ
- ,
- k
-
- a
- λ
- ,
- i
- +
- m
- ,
- j
- +
- n
- ,
- position (i, j) by calculating wherein parameters i, j, m and n are integers, k represents a number of iterations for optimizing the LMS filter, R represents a support region of the LMS filter, wλ,m,nθ,k represents a temporary coefficient of the filter for the subband {λ, θ} at position (m, n) during a kth iterated optimizing procedure, and aλ,i+m,j+n represents an original coefficient of the low subband of level λ at position (i+m, j+n);
- (b) obtaining an error value eλ,i,jθ,k of the high subband {λ, θ} at position (i, j) by subtracting the reconstructed coefficient {circumflex over (d)}λ,i,jθ,k from an original coefficient dλ,i,jθ of the high subband {λ, θ} at position (i, j);
- and
- (c) adjusting the temporary coefficient wλ,m,nθ,k of the LMS filter by adding μeλ,i,jθ,kaλ,i+m,j+n, wherein μ represents an adjusting parameter that used to control a rate of convergence for optimizing the coefficients of the LMS filter, then iterating the steps (a) and (b) until an overall change of the coefficients of the LMS filter is equal to or less than a predetermined value.
Dependencies are inferred from the claim text.