The document proposes an image watermarking algorithm that is robust against geometric attacks. It embeds a watermark image into the medium frequency coefficients of the host image's wavelet domain. It extracts invariant centroids from the watermarked image to correct for geometric transformations like rotation and scaling. Experimental results showed the algorithm can withstand common signal processing and geometric attacks.
1 of 24
Downloaded 44 times
More Related Content
Watermark
2. Abstract
To solve the sensitive problem of signal processing and geometric
distortion of digital image watermarking, an image watermarking
algorithm against geometric attacks was proposed in the paper.
After decomposing the whole image with 3 level of discrete wavelet
transform and transforming the watermark image by Arnold
shuffling, embed the watermark data to the media frequency
coefficients of wavelet domain according to the conceal quality of
Human Visual System (HVS); and extract two invariant centroids as
factors to correcting geometric transformation by using the
theories of invariant centroid, the watermarked image could be
corrected. The experimental results show that the algorithm is
robust to general signal processing and geometric attack such as
rotation, scaling and translation.
3. Secure Part In Secure Part In Secure Part
Result
Embedding Attacking Detecting
Signals
Function E Function A Retrieval Function
S
General digital watermark life-cycle phases with embedding-, attacking-, and
detection and retrieval functions
6. (1) Use Haar wavelet, the images A be done 3 level discrete wavelet
transform, to produce LL3, HL3, LH3, HH3 and so on ten sub-band.
(2) Use ascending order for the intermediate region HL3, LH3 of
image, get the sequence C, and note the location corresponding to
order
(3) Arnold scrambling the watermark information, then obtain
scrambling watermark information W.
(4) Using the multiplicative rule, large absolute value coefficient
with C embed , then get the watermark information W.
c′ i = ci(1 + alpha*wi)
where the size of determines the intensity of the image frequency
modified by the watermark signal.
7. (5) According to corresponding sequence in step (2), the modified
media frequency sequence c′
i is assigned to corresponding location of original intermediate
frequency regions HL3, LH3.
(6) Use the modified wavelet coefficients in step (5) by discrete
inverse wavelet transform to get image embedded with the
watermark information.
(7) Extract the two invariant centroid points tm, tn of the images
embedded watermark information, and obtain the coordinates and
corresponding radius r1, r2 of the two points as geometric
distortion correction key for watermark detection.
8. (1) Use the methods described before as well as the key of geometric distortion of the
watermark image rotation, scaling, translation correction.
(2) Use DWT for watermarking image A∗ with geometric distortion correction to get LL3, HL3,
LH3, HH3 and so on ten sub-bands.
(3) According to the corresponding position sequence and the embedded watermark
sequence size, a embedding position of intermediate frequency regions HL3, LH3 in
watermark
image is determined, and embedded watermark sequence c′ i is obtained.
(4) Use the Eq. (4), to get scrambling watermark information W′.
W′
i = (c′ /ci − 1)/alpha
(5) Use the saved Arnold scrambling key to do periodic transformation for W′, then get the
extracted watermark image W∗.
9. • Extraction of the Invariant Centroid
• Parameter Correction of Geometric Distortion
• Image Rotation Correction Algorithm
• Image Scaling Correction Algorithm
• Image Translation Correction Algorithm
10. Wavelet
Host Image LL3,LH3,HL3
transform
Another strategy use only high value Use LL3 but
coefficients to hide your coefficients low energy or
LH3 and/or HL3
for higher
energy
Another strategy use additive way or
multiplicative way
11. Wavelet Wavelet Wavelet
Host Image transform
transform transform
Watermark
Image LL3,LH3,HL3
Use LL3 but
Arnold low energy
scrambling
Watermarked Wavelet
+*
Image inverse 3 levels