Digital Watermarking Using Phase Digital Watermarking Using Phase Dispersion --- UpdateDispersion --- Update
SIMG 786 SIMG 786
Advanced Digital Image ProcessingAdvanced Digital Image Processing
Mahdi Nezamabadi, Mahdi Nezamabadi,
Chengmeng Liu, Chengmeng Liu,
Michael SuMichael Su
OutlineOutline
• Carrier design
• Embedding and extraction for single tile and Multi-tiles (improving the robustness)
• Parameter α selection and invisibility
• Moment Normalization• Rotation/Scale Detection
Carrier Implementation 1Carrier Implementation 1
• Carrier is implemented in frequency domain
• Carrier has random phase
• The amplitude of Carrier is high pass in order to make it invisible in spatial domain
• Carrier should be symmetric in frequency domain in order to make its imaginary part to 0 in spatial domain
Carrier Implementation 2Carrier Implementation 2High-pass vs. All-passHigh-pass vs. All-pass
Carrier Implementation 3Carrier Implementation 3
• Auto correlation of Carrier function should approximate delta function
• The average of Carrier should be 0
Carrier Implementation 4Carrier Implementation 4if Carrier is not symmetric in frequency domainif Carrier is not symmetric in frequency domain
Embedded MessageEmbedded Message
*• Convolution is
implemented by multiplication of Fourier transform in frequency domain
• Zero padding must be performed before FFT
Tiling Improves the RobustnessTiling Improves the Robustness
Tiling Improves the RobustnessTiling Improves the Robustness
• After 8 by 8 tiling, the summation of tiles is shown at right
• The amplitude of the input image will be averaged to flatten after summation of 64 tiles
• The watermark information is amplified
Parameter Parameter αα = 0.1 = 0.1
αα = 0.05 = 0.05 αα = 0.1 = 0.1 αα = 0.3 = 0.3 αα = 0.5 = 0.5 αα = 0.7 = 0.7
Parameter Parameter αα = 0.3 = 0.3
αα = 0.05 = 0.05 αα = 0.1 = 0.1 αα = 0.3 = 0.3 αα = 0.5 = 0.5 αα = 0.7 = 0.7
Parameter Parameter α = 0.5 = 0.5
αα = 0.05 = 0.05 αα = 0.1 = 0.1 αα = 0.3 = 0.3 αα = 0.5 = 0.5 αα = 0.7 = 0.7
Parameter Parameter α = 0.7 = 0.7
αα = 0.05 = 0.05 αα = 0.1 = 0.1 αα = 0.3 = 0.3 αα = 0.5 = 0.5 αα = 0.7 = 0.7
Similarity vs. Similarity vs. α
• Similarity is measured by cross correlation between original and extracted log
• 64 tiles were used in embedding
• The α controls the visibility of the watermark logo in the watermarked image
• The α also depends on the number of tiles
Attacked by low pass filterAttacked by low pass filter
• The watermarked image is blurred
• The extracted logo is equivalent to original log convolve with a low pass filter
α=0.3,no blurred α=0.3,blurred
Moment NormalizationMoment Normalization
• Preprocessing to remove the high amplitude, low frequency noise
• At flat area, v’ is replaced by random number with variance of σd
Rotation/Scale DetectionRotation/Scale DetectionThreshold and image DilationThreshold and image Dilation
Rotation/Scale DetectionRotation/Scale DetectionImage rotationImage rotation
Rotation/Scale DetectionRotation/Scale DetectionImage rotationImage rotation
Current Issues and ProblemsCurrent Issues and Problems
• Odd and Even dimensions of Carrier function generate different output result in spatial domain.
• How to deal with interpolation errors during rescaling and re-rotation processes
Follow-up WorksFollow-up Works
• Implement Contrast Sensitivity Function in Carrier function design
• Rotation/Scale pattern detection • Rotate back to right orientation and scale back
to its original dimensions• Implementation of Binary Message template
function• Integrate all functions and final presentation and
report
Thank You!Thank You!
Question?Question?