Cryptography Reference
In-Depth Information
a PSNR value higher than 35 dB. Tians algorithm marginally outperforms
Alatters algorithm at lower values of PSNR.
13.4 Our Techniques
We did a great deal of research on lossless data hiding techniques and as a
result we propose a novel reversible embedding method [16] to resist cropping
attack. The method can find the cropping location by embedding the locating
pattern in the original image.
13.4.1 Reversible Watermarking Resisting to Cropping Attack
How to resist some attacks remains a challenging question in reversible wa-
termarking research. We propose a reversible watermarking scheme able to
resist cropping attacks in the spatial domain. The proposed scheme embeds a
locating decoding pattern in host image using the patchwork algorithm and
a modulo 256 addition. The locating decoding pattern is used to retrieve the
cropped positions in the given watermarked image which was possibly cropped
during the extraction process. The locating decoding pattern modifies a small
partition of pixel values in each disjoint image block. An improved difference
expansion (DE) method is proposed to embed a payload known as an au-
thentication hash to the original image block. This introduced into the rest of
the pixel values in each image block. The traditional modulo additions would
cause an annoying visual artifact similar to the salt-and-pepper noise. The
scheme adopts the method of keeping the flipped pixels unchanged to effec-
tively avoid an artifact. Since we use the image ID and the block index as the
input to the hash function, the scheme can e
ciently resist to the collage at-
tack. Experimental results show that the scheme can achieve good robustness
against both cropping attack and collage attack.
13.4.2 The Improved Difference Expansion Method
Tian [11] classifies difference values into three disjoint sets according to the
above definitions. The first set contains all selected expandable differences.
The second set contains all those expandable but which are not in the first
set and all the changeable differences. The third set contains the rest of the
differences which are not changeable. Tian uses a location map to record
the positions of those selected expandable differences in the original image.
The location map is losslessly compressed and the compressed bitstream is
embedded into the host image as an overhead for using the later reconstruction
of the original image. The compressed location map would consume certain
hiding capacity. Our method largely reduces the length of the compressed
bitstream by adopting different classification method.
