Cryptography Reference
In-Depth Information
Invertible, Lossless, Distortion-Free, or Erasable Data Embedding Technique.
The reversible data embedding techniques either have the desirable properties
of a digital watermark, such as perceptible transparency, robustness, security,
and data capacity, or have reversibility.
From the literature, most data embedding techniques cannot be completely
reversed, since embedded distortion due to discarded information, quantiza-
tion and integer rounding at the boundaries of the grayscale range cannot be
removed. For example, by replacing the bits in the bit-planes, that is, least
significant bit-plane (LSB), with watermarking bits, watermarking techniques
discard all replaced bits of the bit-planes. Consequently, the bit-replacement
is clearly lossy. For watermarking techniques, based on the quantization such
as Vector Quantization (VQ), Quantization Index Modulation (QIM), and
so on, quantization error makes that retrieved pixel values mismatch with
original pixel values. Hence, there is little hope to restore the original image
without distortion. For spread spectrum watermarking techniques in Discrete
Cosine Transform (DCT) domain and/or Discrete Wavelet Transform (DWT)
domain, round-off error and truncation error make invertible watermarking
impossible. Additive, non-adaptive schemes (truncation addition) are almost
lossless except for the pixels with grayscales close to 0 or 255 where truncation
has occurred owing to overflow or underflow.
In all of the above mentioned embedding techniques are not suitable in
some applications. Those require high-precision, such as the medical images,
artworks, and so on. Some watermarking techniques have been developed in
order to presented to satisfy the reversible requirement over the last a few
years. All existing reversible watermarking techniques can be classified into
two categories.
Type-I algorithms [1]-[4] are based on robust, spatial additive watermarks
combined with modulo addition. These techniques add the payload by
modulo addition to the host image during embedding process. At the de-
coder, the payload can be reconstructed from the watermarked image,
and then it is subtracted to restore the original image. However, mod-
ulo additions would cause a disturbing visual artifact resembling a corre-
lated salt-and-pepper noise into the watermarked image when pixel values
close to the maximally allowed value are flipped to zero and vice versa.
Type-I algorithms generally combine statistical approaches. For example,
the patchwork algorithm with modulo additions is used to ensure correct
watermark extraction. Hence, Type-I algorithms are robust for the data
embedding and allow for extraction of hidden data even for the perturbed
watermarked image. It can not ensure that the original image is precisely
retrieved. The algorithm in [4] has certain degree of robustness against
JPEG lossy compression. This is the only existing robust lossless data
embedding algorithm for use against JPEG compression.
