Convolutional Neural Network Architecture for Recovering Watermark Synchronization

. 2020 Sep 22;20(18):5427. doi: 10.3390/s20185427

Algorithm 1 QIM-based watermark embedding procedure.

1:
Input: curvelet coefficients $C^{s, l} (i, j)$ over scale s and direction l
2:
Output: modified curvelet coefficients $C_{m}^{s, l} (i, j)$
3:
q: quantization step
4:
b: bit to be inserted
5:
6:
$A^{s, l} = \frac{1}{m n} \sum_{i = 1}^{m} \sum_{j = 1}^{n} a b s (C^{s, l} (i, j))$
7:
if mod[round( $A^{s, l} / q), 2]$ = b then
8:
$c = 0$
9:
else
10:
if mod( $a b s (A^{s, l}) / q, 1) \leq 0.5$ then
11:
$c = 0.5$
12:
else if mod( $a b s (A^{s, l}) / q, 1) > 0.5$ then
13:
$c = - 0.5$
14:
end if
15:
end if
16:
$Q^{s, l} = (round (a b s (A^{s, l}) / q + c) \cdot q) / A^{s, l}$
17:
$C_{m}^{s, l} (i, j) = C^{s, l} (i, j) \cdot Q^{s, l}$