Cryptography Reference
In-Depth Information
13.4.3 The Locating Decoding Pattern
General scheme for reversible watermarking is shown in Fig. 13.10. The host
image is partitioned into not-overlapping blocks with size NN starting
from top-left corner. All disjoint image blocks are denoted as B
k
(x, y), where
1≤x, y≤N, 1≤k≤n,andn is the total number of blocks in the
image. The locating decoding pattern is the size of the whole image, created
by tiling, using the same basic pattern (block) of size NN .Weuseaone-to-
one correspondence between each image block and the basic decoding pattern
in x and y coordinates.
We assume that the value at each point in the locating decoding pattern
is initialized to zero. Independently in the basic decoding pattern, we utilize
the same random generator to determine a small partition of all N
2
points.
There are a total of 100 points. These are divided into two equal sets of points,
A and B. A given intensity correspond to δ value used below is respectively
added to and subtracted from the values at all the points of the patches A and
B. We add a small constant value δ to the values at all the points of patch
and subtract the same value δ from the values at all the points of patch B.
Hence, the values of all unchanged points remain zero, while the values of the
changed points are δ or−δ.
Some pixel values in the k-th image block, B
k
(x, y), having the same x and
y coordinates as the points incremented or decremented by the δ value in the
basic decoding pattern, correspond to add or subtract a constant value δ. This
is done according to modulo 256 addition. Modulo additions would cause an
annoying visual artifact resembling a correlated salt-and-pepper noise when
the pixel values close to the maximally allowed value are flipped to zero and
vice versa. To avoid the salt-and-pepper noise, we keep those flipping pixel
values (p∈([0, 3]∪[252, 255]), δ = 4), p is one pixel value) unchanged. We
create a binary location map called as pM ap which contains the position infor-
mation for all the changed pixels in the patches A and B. Symbol 1 indicates
unchanged pixels and the symbol 0 denotes changed pixels of patches A and
B in the pM ap. The size of this is equal to the number of patches A and B.
13.4.4 A Watermarking Embedding Scheme
Based on the Improved Difference Expansion Method
The embedding process is done in two stages. Stage 1 embeds a locating
decoding pattern according to the method described above in Section 13.3.3.
The Stage 2, the embedding process is shown in Fig. 13.11. In each block,
for the remnant pixels-instead of pixels selected for the locating decoding
pattern, we employ an improved DE method to embed a payload. The remnant
pixels are grouped into pairs of pixel values. A pair consists of two neighboring
pixel values or two values with a small difference value. We apply the integer
transform in Eq. (13.12) to each pair.
