Professional Documents
Culture Documents
Encoder
Compressed
image data
Decoder
Entropy
• In information theory, entropy is a measure of the
uncertainty in a random variable. The term usually refers to
the Shannon entropy, which quantifies the expected value
of the information contained in a message.
• A fair coin has an entropy of one bit. A series of two fair
coin tosses has an entropy of two bits. The number of fair
coin tosses is its entropy in bits.
14
Bit Plane Coding
It is possible to see that the first bit plane gives
the roughest but the most critical
approximation of values of a medium, and the
higher the number of the bit plane, the less is
its contribution to the final stage. Thus, adding
a bit plane gives a better approximation.
If a bit on the nth bit plane on an m-bit dataset
is set to 1, it contributes a value of 2(m-n),
otherwise it contributes nothing.
15
For example, in the 8-bit value 10110101 (181 in decimal) the
bit planes work as follows:
Bit Plane Value Contribution Running
Total
1st 1 1 * 2^7 = 128 128
2nd 0 0 * 2^6 =0 128
3rd 1 1 * 2^5 = 32 160
4th 1 1 * 2^4 = 16 176
5th 0 0 * 2^3 = 0 176
6th 1 1 * 2^2 = 4 180
7th 0 0 * 2^1 = 0 180
16
8th 1 1 * 2^0 = 1 181
17
Transform Coding
A reversible linear transform (such as Fourier
Transform) is used to map the image into a set of
transform coefficients
These coefficients are then quantized and coded.
The goal of transform coding is to decorrelate pixels
and pack as much information into small number of
transform coefficients.
Compression is achieved during quantization not
during the transform step
Transform Coding
Software Research
Wavelet Coding
Wavelet Coding
High Frequency coefficients tend to be very
small --- 0
24
Outline
Introduction
State-of-the-Art
Characteristics of Visible & Invisible
Watermarking Schemes
Attacking Visible & Invisible Watermarking
Schemes
Discussions and Conclusions
Classifying Watermarking Schemes
Data hiding
Steganography Watermarking
+
Watermark
Imperceptible distortion Visibly Meaningful
Perceptibility
pattern
I ' K 1 * I K 2 *W
D ( E I ( I ' ), E I ( I )) Threshold I
D ( EW ( I ' ), EW (W )) Threshold W
I’: the watermarked content
I: the un-watermarked original content
W: the watermark pattern
Ki: the weighting factor
D: a distance function measuring the perceptual difference of its two
parameters
Ei: image feature extraction operators
ThresholdI: the largest allowable distortion of image details that observers
can tolerate and, at the same time, the signature of can be maintained.
ThresholdW: the largest allowable distortion of the embedded watermark
pattern that the copyright information can be clearly recognized.
Other Enhancing Schemes
[Meng and Chang]
The same embedding model is extended to the DCT domain by simple
statistic model approximation for the convenience of processing directly
in the MPEG-compressed domain.
[Kankanhalli et al]
Local features related to the degree of distortion tolerances, such as
edge locations, texture distributions and luminance sensitivity, are taken
into consideration so that more unobtrusive watermarked images can be
generated.
Simple statistics of block-DCT coefficients are calculated and analyzed
to decide the watermark embedding energy of each block.
Edge integrity will be preserved, in these approaches, since the
edge information is essential to maintain the image quality.
And the energy of the embedded watermark is larger in highly
textured areas than in smooth ones due to different noise
sensitivity.
In additions, the watermark energy of mid-gray regions is also
smaller than other areas since the noises are more visible against a
mid-gray background
[S. P. Mohanty et al]
in addition to the visibly embedded watermark, a fragile invisible
watermark is also adopted to check if the visible watermark is altered or
Important observations (1/4)
Image inpainting
is an iterative image recovery technique.
successfully reconstruct the edges of damaged area.
Basic Inpainting Attacks
Edge Area
Inexpensive copies
No quality loss
Wide distribution but no control
45
The use of digital watermarks
(Data Hiding )
46
What is watermarking?
47
Requirements
Requirements vary with application. For
example:
Perceptually transparent - must not
perceptually degrade original content.
Robust - survive accidental or malicious
attempts at removal.
Oblivious or Non-oblivious - Recoverable
with or without access to original.
Capacity – Number of watermark bits
embedded.
Efficient encoding and/or decoding.
Example: Additive Watermarks
I(x,y) IW (x,y)
Multiply by gain
k factor k IW(x,y)= I(x,y)+ k W(x,y)
fv
fh
Watermark Attacks
Active Attacks.
Hacker attempts to remove or destroy the
watermark.
Watermark detector unable to detect
watermark.
Key issue in proof of ownership, fingerprinting,
copy control.
Not serious for authentication or covert
communication.
Covert Communication
Forgery Attacks.
Hacker tries to embed a valid watermark.
Serious in authentication.
If hacker embeds a valid authentication
watermark, watermark detector can accept
bogus or modified media.
Example: Forging Photo I.D.
Millionaire
with Swiss
bank Bad
account Forgery
!
Thanks
Hong Man,
for this
picture!
Assistant
Professor
Content-based Watermarking
Original Watermark
JPEG Compression Attack
Watermarking Research
Information Theoretic Issues.
Compute “watermarking capacity”
How many watermark bits can be hidden in
a watermark channel ?
Capacity when multiple watermarks are
used.
Decision Theoretic Issues.
Robust watermark detector.
Quickest possible watermark detection.
Invariant to statistical uncertainties.
Achieves desired false alarm and detection
probabilities.
Watermarking Research
Signal Processing Issues.
What transforms to use? – wavelet, DCT etc.
Geometric invariant transform – scale,
translation, and rotation invariance.
Watermarking protocols and system
issues.
Protocol used by sender and receiver for
secure transmission.
Steganalysis.
How to analyze multimedia data for the
presence of watermark(s).
Robust Watermarking Techniques -
A Communications Framework
Noise/Attack Channel
Information Theoretic Formulation
A new watermark channel model.
Multiplicative and additive attacks.
Models geometric attacks (“killer” attack).
Attack parameters assumed imperfectly
known.
Lower and upper bounds on watermarking
capacity.
Many current capacity estimates are specific
cases of proposed estimates.
Capacity in the presence of multiple
watermarks using multiuser information theory.
Multiple Description Watermarking
70
Watermarking process
Extraction stage (3
Blind – extraction without original image
Semi-blind – rely on some data or features
Non-blind – need original image
Detection stage (4
Evaluate the similarity between the original and
detected watermark
False positive:watermark is detected although there is
none
False negative:no watermark is detected while there is
one
Watermarking properties
Perceptual transparency
Robustness
The mark should resist to
Common signal processing like lossy compression
Geometric transformation like image rotation, scaling, and
cropping
Security
How easy it is to intentionally remove a watermark
Data capacity
Amount of information that can be stored within the
content
Categories of digital watermark
Perceptible
Imperceptible
Robust
ownership assertion
Fragile
indicate modifications of the content
Semi-fragile
differentiate between lossy transformation that are
“info. preserving” and lossy transformation which are
“info. altering”
Applications
Copyright Protection
Invisible watermark which can tolerate malicious and
unintentional attacks
It does not prevent people from copying the digital
data
Data Hiding
It tries to invisibly embed the maximum amount of
data into a host signal => this allows communication
using enciphered messages without attracting the
attention of a third party
Robustness is not important while invisibility and
capacity are required
Applications
Authentication and Data Integrity
Verification watermarks are required to be fragile, so
that any modification to the image will destroy the
mark
Copy Protection
Requirements
Robustness against removal
Ability of blind detection
Capability of conveying non-trivial number of bits
Examples of watermarking
Image Watermarking
Document Watermarking
Graphics watermarking
Video watermarking
Image Watermarking
Robust and imperceptible
The watermark may be scaled appropriately to
minimize noticeable distortion to the host
Examples
Texture-based watermark
Embed it into a portion of the image with similar texture
Insert a watermark into the phase components
The phase information is perceptually more important than
the magnitude data
Document Watermarking
Line-shifts
Word-shifts
Document Watermarking
Slight modifications to characters
Graphics Watermarking
Embedding in facial animation parameter (FAP)
data
The amount of deviation the watermark signal has on
FAP is limited to minimize visible distortion
1% for head rotation
3% for lip motion
Video Watermarking
Copy generation management
Minimum information that must convey
Copy never
Copy once
Copy no more
Copy freely
For DVD, real-time decoding is required
Detector placement
Detection in the drive
Detection within the application (within MPEG
decoder)
Video Watermarking
Examples
For video coding like MPEG or H.26x, we embed the
watermark into DCT coefficients
Only partial decoding of block DCT is necessary for watermark
embedding
If constant bit rate is required, only nonzero DCT coefficients are
marked
Layered Access Control Schemes
on Watermarked Scalable Media
Scalability – spatial, temporal and SNR
scalability
A typical MPEG-2 conditional access receiver
Layered Access Control Schemes
on Watermarked Scalable Media
A typical MPEG-2 conditional access receiver
Control Word – change very often so that fast (simple)
decryption algorithm would not be broken easily
EMM (Entitlement Management Message) – with the
aid of EMM, the broadcaster can change the status of
the user accessibility of contents
ECM (Entitlement Control Message)
Layered Access Control Schemes
on Watermarked Scalable Media
A typical MPEG-2 conditional access receiver
Problem
We send the key to user via reliable channel, while the
content goes through an unreliable channel.
Key and contents won’t arrive at the same time
=> Synchronization problem
Solution
Using robust watermark to embed the ECM and EMM
information into the content
Existing methods
Method in the spatial domain and in the
frequency domain.
Visible watermarking systems are usually
able to sustain all possible image
alterations and even intentional
disturbances.
87
Fingerprinting or traitor tracing
Marking copies of one document with a customer signature.
original
+
…
W1 W2 WN
… N customers
Robust, secure, invisible watermark, resistant with respect
to the collusion attack (averaging copies of documents with
different marks).
Adding captions to images, additional
information to videos
Typical application:
Adding subtitles in multiple languages•
Additional audio tracks to video•
Tracking the use of the data (history file)•
Adding comments, captions to images•
Watermark requirements:
Moderately robust scheme•
Robustness with respect to lossy compression, noise adding,•
and A/D D/A conversion
Original images (frames) not available for message extraction•
Security requirement not so strong •
Fast detection, watermark embedding can be more time •
consuming
Watermarking principles
In spatial domain In transform domain
watermark embedded watermark embedded in the
by directly modifying transform space by modifying
the pixel values coefficients
+ =
DCT Inverse DCT
Modify
DCT
Watermarking for color images
One or more selected color channels. •
Luminance•
Approaches:
Watermarking by small blocks (good for cropping)•
Embedding patterns with known geometry•
Watermarking using Fourier-Mellin transform (scaling and •
rotation converted to shift)
Embedding watermarks into image features or salient points•
Weak points:
Computational complexity•
More powerful geometric attacks - StirMark•
Forensic analysis
Localized analysis of
- noise
- histogram
- colors
Break easily
Properties: Computationally cheap
Good localization properties
Too sensitive for redundant data
Examples:
Embedding check-sums in the LSBs
Adding m-sequences to image blocks
Robust watermarks on small blocks
Medium robustness
Insensitive to small changes
Properties:
Not as good localization properties
Can distinguish malicious and
non-malicious modifications
64 pixels
Robust
bit extractor
B 50 bits
Watermarked
Secret W(K, B) B block B
key K Synthesizing
Gaussian + =
Block # sequence
B
Hybrid watermark
Examples:
Robust watermarks on medium blocks
combined with a fragile watermark
Images with Selfcorrecting Capabilities
Tampered image - The license plate The original license plate after
has been replaced with a different one reconstruction
Secret key
Oblivious secure watermarking
Possible approaches:
Content Locked Coordinate Systems •
Feature-based techniques•
Problems with DCT
A given image cannot be queried for
ownership without the original un-
watermarked image (non- oblivious).
Robustness.
It is known that the wavelet image/video coding,
are included in the image/video compression
standards, such as JPEG 2000 and MPEG4 due
to excellent performance in compression.
Wavelet transforms set high frequency
components = 0 (Transients eliminated).
Therefore, it is important to study water-marking
methods in the wavelet transform domain. 103
Watermarking in the DWT domain
DWT has a key advantage it has over
Fourier transforms is temporal resolution: it
captures both frequency and location
information (location in time).
includes two parts:
encoding
Adding the watermark to the original image.
decoding.
Recognizing or extracting the watermark.
104
The DWT
105
DWT pyramid decomposition
An image can be
decomposed into a
pyramid structure with
various band information
.
such as: HH,LH ,LL and
HL frequency bands.
106
Encoding and decoding scheme
1. De-compose an image into several bands with
a pyramid structure.
2. Add the watermark message.
3. Then, we take the two dimensional IDWT of
the modified DWT coefficients.
4. The decoding will be done by applying the
inverse procedure.
107
What we are going to see?
111
Encoding-
4. For the resultant image to fit within the 0
to 255 integer values:
112
First method –Decoding
DWT HL1
-
LH1 HH1 cross
HH1
Watermarked Image correlation
with the
watermark
DWT HL1
is there a
LH1 HH1
HH1 peak?
Original Image
YES NO
113
STOP
First method –Decoding
DWT HL1
-
LH1 HH1 cross
Watermarked Image correlation
with the
watermark
DWT HL1
is there a
LH1 HH1
peak?
Original Image
YES NO
114
STOP
First method -Decoding
119
Second method- decoding
122
Example
123
The third method-overview
The algorithm in WaveMark uses discrete
wavelet transforms and error-correcting
coding schemes to provide robust
watermarking of digital images.
The watermark recovery procedure does not
require a match with an uncorrupted original
image.
The system is practical for real-world
applications, encoding or decoding images at
the speed of less than one second each on a
Pentium Pro PC.
124
Using wavelet
The use of Daubechies' advanced
wavelets makes the watermarked
images more perceptively faithful than
Haar.
The watermark is adaptively applied to:
Different frequency bands
Different areas,based on the smoothness
increases robustness within the limits of
perception.
125
Encoding
126
Encoding
127
Error correction coding
A Hamming code is used to add
redundancy to the bits so that the errors
can be detected or corrected to a certain
extent.
we use a (8,4) extended Hamming code.
4 bytes are the input
8 bytes include the input +4 parity bytes.
Such a code can correct a one-bit error
and detect up to three one-bit errors.
128
Getting the final result
After the watermark code is placed in the
transform matrices, we perform a 4-level inverse
wavelet transform for each of the three matrices
using Daubechies-4 wavelet.