Bit-Plane Watermarking for SPIHT-Coded Images 台北科技大學資工所 指導教授:楊士萱 學生:廖武傑 2003/07/29
OUTLINE Introduction to watermarking SPIHT Factors that affect watermarking system Proposed watermarking scheme Conclusion
Watermarking The watermark is an imperceptible but indelible code used for ownership identification. Why? convenience of digital multimedia data intellectual property of digitally recorded material shortcomings of encryption techniques Requirements perceptual transparency robustness security appropriate complexity Computational requirements introduced by watermark embedding and detection should be small
Domains of image watermarking
OUTLINE Introduction to watermarking SPIHT Factors that affect watermarking system Proposed watermarking scheme Conclusion
SPIHT coding procedure
Wavelet transform(1D) h0 h1 h0 h Resolution 1/2 Resolution 1/4 C 3,n C 2,n C 1,n d 2,n d 1,n
Wavelet transform(2D) h0 h Horizontal filteringVertical filtering
Wavelet filter Daubechies 9/7 Analysis Filter Coefficients: iLow-Pass FilterHigh-Pass Filter ± ± ± ±
Example, “Lena”
SPIHT coding procedure
Spatial Orientation Trees
Example (1 st Sorting pass, Encoder) LIPLSP LIS: T=32 A2A3A
Example (1 st Sorting pass) LIPLSP LIS: T=32 A3A B
Example (1 st Sorting pass) LIPLSP LIS: T=32 A B B3 001
Example (1 st Sorting pass) LIPLSP LIS: T=32 A B A9 001 A10A11A
Example (1 st Sorting pass) LIPLSP LIS: T=32 A B A9 001 A11A
Example (1 st Refinement pass) Do nothing
Example (2nd Sorting pass) LIPLSP LIS: T=16 A4B2A9A11A
Example (2nd Refinement pass) LSP: In 1st sorting pass:
Example (3rd Sorting pass) LIPLSPT=8
Example (4th Sorting pass) LIPLSPT=4
Example (5th Sorting pass) LIPLSPT=2
Example (1 st Sorting pass, Decoder) Received: originaldecoded
Example (1 st Sorting pass, Decoder) Original coefficients Reconstructed value (by 1st sorting pass) 48
Example (2 nd Sorting pass, Decoder) Received: originaldecoded
Example (2 nd Refinement pass, Decoder) Original coefficients Reconstructed value (by 1st sorting pass) Refinement bit: 56 Refinement by 2nd refinement pass 40 If refinement bit is ‘0’
Example (2 nd Refinement pass, Decoder) Received: originaldecoded 1010
OUTLINE Introduction to watermarking SPIHT Factors that affect watermarking system Proposed watermarking scheme Conclusion
Watermarks over a noise channel The noise comes from: The host signal Lossy compression Other attacks
Host-interference-nonrejecting watermarking method Transform domain watermarking Watermark: Gaussian random source Embedding method: After distortion: Effect of host interference and quantization noise W. Zhu, Z. Xiong, and Y.-Q. Zhang, “Mutiresolution watermarking for images and video,” IEEE Trans. Circuits Syst. Video Technol., June 1999
Watermark similarity measure for judging: (linear correlation) where: Informed detection: Effect of host interference and quantization noise (cont.)
Test images “Lena”“Baboon”
Degradation of watermark similarity due to quantization Bpp1/641/321/161/81/41/21 Step size MSE PSNR Correlati on (informe d) Correlati on (blind) “Lena”, alpha=0.26, watermark length:255, subband:1-6, coefficients smaller than 512
Degradation of watermark similarity due to quantization “Baboon”, alpha=0.28, watermark length:255, subband:1-6, coefficients smaller than 512 Bpp1/641/321/161/81/41/21 Step size MSE PSNR Correlati on (infor med) Correlati on (blin d)
OUTLINE Introduction to watermarking SPIHT Factors that affect watermarking system Proposed watermarking scheme Conclusion
Proposed watermarking scheme
Proposed watermarking scheme (Cont.) Embedding our watermark
The effect of wavelet coefficient after embedding watermark
Extracting watermark procedure
Watermarking detecting
Proposed watermarking scheme (Cont.) Host-interference rejection Minimization of quantization effects Low complexity Progressive nature
StirMark attacks JPEG compression Gaussian filtering Median filtering Sharpening FMLR
JPEG compression attacks
Gaussian filtering
Median filtering
Sharpening
FMLR
watermark length:255 PN code, subband:1-6, coefficient between 512 and 256 soft-decision, bit-rate: 0.25bpp LenaBaboonPepperF16 Jpeg 50% PSNR / / / / / / / Jpeg 30% PSNR / / / / / / / / Jpeg 10% PSNR / / / / / / / / Gaussian filter PSNR / / / / / / / / Median filter PSNR / / / / / / / / Sharpening PSNR / / / / / / / / FMLR PSNR / / / / / / / /
watermark length:255 PN code, subband:1-6, coefficient between 512 and 256, soft-decision, bit-rate: 0.5bpp LenaBaboonPepperF16 Jpeg 50% PSNR / / / / / / / / Jpeg 30% PSNR / / / / / / / / Jpeg 10% PSNR / / / / / / / / Gaussian filter PSNR / / / / / / / / Median filter PSNR / / / / / / / / Sharpening PSNR / / / / / / / / FMLR PSNR / / / / / / / /
Enhanced watermarking by repetition codes
Performance (basic vs. enhanced) By soft-decision: The difference between basic and enhanced scheme is small. By hard-decision: Enhanced scheme is superior to basic scheme. (except some particular attacks)
OUTLINE Introduction to watermarking SPIHT Factors that affect watermarking system Proposed watermarking scheme Conclusion
Our watermarking system suffers less from the host interference and quantization noise. The embedding and extracting of watermarks are simple. The performance can be improved by enhanced scheme with larger noise margins.