Improved RobustWatermarking Based on Rational Dither Modulation - Advances in Image and Graphics Technologies - page 309

3 Proposed Watermarking Methods

We implement our improved method in wavelet domain. The lowest frequency sub-band
is selected, because it is the perceptually significant region which is robust enough to
resist various attacks. The following are the details.
1)Preprocessing: The significant coefficients, which have large magnitude, are cho-
sen to embed watermark, because they are more robust and the allowed modification
strength of them is larger, which makes the embedding more robust to attacks. In our
scheme, we divide the lowest frequency sub-band into non-overlapping blocks, and se-
lect the largest two coefficients from each block to embed one bit. However, in natural
images, some parts do not contain any significant coefficient, which are improper for
embedding, or some parts contain more than two significant coefficients, which may
cause a waste. In other words, the significant coefficients are not uniformly distributed.
To settle this problem, before partitioning the lowest sub-band into blocks, we first
scramble the selected sub-band, so that the order of the coefficients will be disrupted
and significant coefficients are distributed more uniformly.
2) Watermark embedding: The watermark embedding procedure is illustrated in Fig.
1, which can be described as following steps:

1. D level DWT is applied on the host image.
2. The lowest frequency sub-band is selected and scrambled using a secret key K .
3. Divide the scrambled sub-band into non-overlapping blocks with size of w .
4. Select the largest two coefficients from each block, then quantize the ratio of them
to embed a watermark bit as introduced in Section 2.2.
5. Finally, the scrambled sub-band is descrambled and inverse discrete wavelet trans-
form is applied, then the watermarked image is generated.

Partition LL n
into blocks

DWT LL n K Scrambling

Host Image DWT LL n K Scrambling

Embed the watermark
by quantizing the ratio

Select two significant
coefficients from each block

Select two significant
coefficients from each block

Partition LL n
into blocks

Watermark

Watermark
Decoder

Extracted
Watermark

Watermarked
Image

Descrambling IDWT

Fig. 1. Flowchart of watermark embedding

Fig. 2. Flowchart of watermark decoding

3)Watermark decoding: The process of watermark decoding is illustrated in Fig. 2,
which can be described as follows:

1. The watermarked image is decomposed with D level discrete wavelet transform,
then the lowest frequency sub-band is scrambled with the secret key K and divided
into non-overlapping blocks with size of w , as described in the first three steps in
the watermark embedding process.
2. Select the largest two coefficients from each block, denoted as z 1 and z 2 .Then
decode the watermarked bit by Equation (12).