2008/12/17. RST invariant digital image watermarking & digital watermarking based audiovisual quality evaluation. Outline

Transcription

1 //7 RST invariant digital image watermarking & digital watermarking based audiovisual quality evaluation Outline Digital watermarking RST invariant image watermarking Audiovisual quality evaluation based on watermarking Jiying Zhao, Ph.D. School of Information Technology & Engineering Introduction to digital watermarking Digital watermarking What is digital watermarking Host signals Applications of watermarking How it works Requirements for watermarking Some examples Digital Watermarking technology allows users to embed some data into digital contents such as still image, movie and audio data. When data is embedded, it is not written at header part but embedded directly into digital media itself by changing media contents data. Host signals for digital watermarking Image Video Audio Software Applications of digital watermarking Rights management Copyrights protection Content distribution, tracking and monitoring Authentication Assure contents integrity Prevent unauthorized alternation of contents Detect alternation location in the contents Contents management Indexing & retrieval Access/copy control Prevent unauthorized copy, playback of multimedia contents Quality evaluation

2 //7 How it works (IBM TRL) Technical requirements for watermarking Invisibility Robustness Capacity Examples of watermarking Broadcast monitoring (example: IBM TRL) Original Watermarked Owner identification Content authentication An invisible watermark is very resistant to destruction under any image manipulation. Verifies ownership of media suspected of misappropriation. Digital detection of the watermark would indicate the source of the image. A fragile watermark will be destroyed when the image is manipulated digitally in any way. It proves authenticity of media. If the watermark is still intact, then the image has not been doctored. If the watermark has been destroyed, then the image has been tampered with. Semi-fragile watermark, tell-tale watermark

3 //7 RST invariant image watermarking [][][] Rotation, Scaling, and Translation (RST) are the most challenging attacks to image watermarking. We have proposed several novel RST invariant image watermarking algorithms. An image rectification scheme and its applications in RST invariant digital image watermarking Fourier transform f ( x, y) F( u, v) Fourier transform shift invariance f ( x a, y b) F( u, v)exp( j ( au bv)) N Fourier transform -- rotation F r f r ( x, y) f (( xcos a ysin a),( xsin a ycos a)) ( u, v) F(( u cos a vsin a),( u sin a vcos a)) Fourier transform -- scaling u v f ( ax, by) F(, ) ab a b Log-polar mapping Using log-polar coordinates: u e cos v e sin The magnitude of the Fourier spectrum: I'(, ) I( ln, ) Advantage: Image scaling results in a translational shift of lnσ along the log-radius ρ axis. Image rotation results in a cyclical shift of α along the angle θ axis. Image translation has no effects in LPM domain.

4 //7 Log-polar mapping (example) Matching template We cut a small block as a matching template from the LPM domain or the spatial domain of the original image. Original image DFT LPM Cut a template optional Cut a template DFT: discrete Fourier transform LPM: log-polar mapping y x Cartesian domain LPM domain The key technique is to match the template in the log-polar domain with the watermarked image having undergone RST attacks. Five traditional filters The matching results of the five filters (without rotation or scaling) Classical matched filter G jg (, ) (, ) (, ) A G e Amplitude-only filter G, ) A (, ) A ( G Inverse filter j (, ) e G GI (, ) A (, ) Phase-only filter G jg (, ) (, ) e G Binary phase-only filter G BPOF e jbpof (, ) (, ) where, Gr BPOF(, ), Gr Gr stands for the real part of the Fourier transform G(, ) (a) Classical matched filter (b) Amplitude-only filter (c) Inv erse filter (d) Phase-only filter (e) Binary phase-only filter The matching results of the five filters (with rotation and scaling) (g) Classical matched filter (h) Amplitude-only filter (i) Inverse filter Phase-only filtering method Phase information is more important than the amplitude information in preserving the visual intelligibility. Correlation detection is only optimal in the case that the signal can be modeled as additive white Gaussian noise. Using only the phase information of the matching template and the LPM spectrum of the watermarked image undergone RST transformations. * r IFFT[ F (, ) G (, )] where jf (, ) (, ) e F G jg (, ) (, ) e (j) Phase-only filter (k) Binary phase-only filter

5 //7 The matching results of our method Rectification (rotation and scaling parameters) (a) Without rotation or scaling (b) With rotation and scaling Compute the cross-correlation between the template and the watermarked image in the LPM domain. Suppose the coordinates of the peak is (, ). We know the original position of the template is (, ) The translations in the LPM domain are: The rotation and scaling parameters in the spatial domain are: ' N' ln( rmax ) M ' ' e Scaling without rotation (a) Rotated by 5 degrees counterclockwise without scaling (c) Scaled by.7 (e) Rotated by 5 degrees counterclockwise after being scaled by.7 Application : Spatial domain Application : Fourier domain Application : LPM domain (b) Correlation with (a). The peak is (d) Correlation with (c). The peak is at (5,7). at (5,). (f) Correlation with (e). The peak is at (,). * The original template position is (5,7) Rotation with cropping Scaling and rotation 5

6 //7 JPEG compression Image quality evaluation based on watermarking [7] Limitation of existing quality metrics A common limitation of the widely used quality metrics, such as PSNR, wpsnr and Watson model: need access to the original work All of them are based on point-to-point difference calculation between the original work and degraded work in spatial domain or in frequency domain This definitely results in big inconveniences at the receiver side in a communication system when the original work is real-time signals such as TV signals Aim of developing watermarking based quality metric To find a quality estimation method which can accurately estimate image/video qualities without the access to the original image/video Digital watermarking based metric can be a good candidate since it can evaluate the image/video quality by simply evaluating the degradation of the extracted watermark Watermark is much smaller than the original work, which is easier to transmit and makes it possible for real-time quality evaluation Main idea of developing watermarkingbased quality metric Use the degradation of watermark to evaluate the quality degradation of the cover work The watermark is invisibly embedded throughout the cover work The watermark is inseparable from the cover work, which means the watermark will undergo the same transformations and distortions as the cover work The watermark can be embedded with suitable vulnerability so that the degradation of the watermark can reflect the quality degradation of the cover work Main idea of developing watermarking based quality metric The proposed scheme is based on DWT The different watermark embedding vulnerability for different cover work Watermark embedding bits Middle frequency High/low frequency Watermark embedding strength Need to: Find a balance between robustness and fidelity Improve accuracy of the results

7 JND JND //7 Automatic control system designed for the watermarking scheme The proposed watermarking-based image quality evaluation scheme Feedforward control Original image Empirical information Feedforward control Initial quantization parameters -level DWT Watermark embedder Adjusted quantization parameters Quantization parameter adjuster Feedback control Controller Emprical watermark portion -level IDWT Increment/Decrement Reference input + Controller Implementer Controlled process Measured output Watermarked image distortion The quantization parameter(s) needed to be adjusted - Disturbances Degraded watermarked image The most contributing block(s) Sensor Original watermark -level DWT Classic quality metric Gradient calculator Watermark extractor Calculated quality + N + Estimation error Acceptable? - Ideal curve Mapping Estimated quality Y Finished Feedback control Quantization method employed by the watermark embedder The watermark extraction and quality measurement Original position before embedding Final position after embedding Degraded watermarked image -level DWT Quantization coefficients Quantization parameter Watermark extractor Watermark extraction A B 5 Original watermark Q( e) DWT coefficien t even quantization parameter DWT coefficien t odd quantization parameter Ideal curve Mapping Estimated quality Quality evaluation The watermarking based quality measurement The watermarking based quality measurement -- ideal curve & mapping The image/video quality will be estimated by evaluating the degradation of the watermark (True Detection Rate) for the watermark degradation Number of successfully detected watermark bits Number of watermark bitsused for embedding Linear interpolation: T(j) Q(j) TC Q E T(j+) Q(j+) TC T( j) QE Q ( j) Q ( j) Q ( j ) T( j) T( j ) Mean Absolute Error (MAE) MAE N i Q ( i) Q ( i) C N E 7

8 Calculated wpsnr (db) Calculated PSNR (db) Calculated PSNR (db) Calculated wpsnr (db) //7 The principle behind the adjustment of watermark vulnerability Evaluation --- Original watermark JND Ideal curve JND Ideal curve Upper bounds P Adjusted result Tested result Lower bounds The by original watermark Tested curve B A C D E The estimation errors on the sample distortions were reduced at the same time Ideal curve used for estimating PSNR Experiment result estimating PSNR Estimated PSNR (db) Estimated PSNR (db) MAE =.7 db MAE =.777 db Ideal curve for estimating wpsnr Experiment result estimating wpsnr Estimiated wpsnr (db) Estimiated wpsnr (db) MAE =.9 db MAE =.797 db

9 JND Calculated JND Calcualted JND //7 Ideal curve for estimating Watson JND Experiment result estimating Watson JND Estimated JND MAE=.5JND For JND, MAE =.57 JND Estimated JND MAE =.5JND Speech quality evaluation based on watermarking [] Watermarking scheme Discrete wavelet transform Watermark bit embedding Low Pass Signal Filters High Pass Speech DWT Coefficient Levels Level Level Blocking Block Block Coefficient Selection Select first 5 largest coefficients to embed watermark Level m Block n ca cd Quantization -- Embed one bit ( or ) on each coefficient Low Pass Filters High Pass ca cd 9

10 ITU-T (P.) PESQMOS ITU -T (P.) PESQMOS ITU-T (P.) PESQMOS ITU-T (P.) PESQMOS //7 Experimental results Experimental results DWMOS DWMOS DWMOS DWMOS (a) MP compression () samples (b) Gaussian noise () samples (c) Low-pass filtering () samples (d) Packet loss () samples DWMOS --- predication of MOS DWMOS --- predication of MOS Research in process The watermarking-based image quality algorithm is applying to the MPEG- / H. video codec. The research on quality estimation against noise pollution, packet loss, and other signal processing are in process. The experiment to simulate subjective visual quality is under consideration. NSERC Strategic Project: Quality evaluation and enhancement of audiovisual signals based on digital watermarking Selected Journal Papers (RST invariant image watermarking). D.Zheng, S.Wang, and J.Zhao, An RST invariant image watermarking algorithm with mathematical modeling and analysis, to appear in IEEE Transactions on Image Processing.. D.Zheng, Y.Liu, J.Zhao and Abdulmotaleb El Saddik, A Survey of RST Invariant Image Watermarking Algorithms, ACM Computing Surveys. Vol. 9, No., Article 5, pp. -9, June 7.. Y.Liu, D.Zheng, and J.Zhao, An image rectification scheme and its applications in RST invariant digital image watermarking, Kluwer Journal: Multimedia Tools and Applications, Vol., No., pp. 57-, July 7.. D.Zheng, Y.Liu, and J.Zhao, RST invariant digital image watermarking based on a new phase-only filtering method, Elsevier Journal: Signal Processing, Vol.5, No., pp.5-7, December Y.Liu, D.Zheng, and J.Zhao, A Rectification Scheme for RST Invariant Image Watermarking, IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences, Special Section on Cryptography and Information Security, Vol. E-A, No., pp. -, January 5, LETTER.. D.Zheng, J.Zhao, and A.El Saddik, RST Invariant Digital Image Watermarking Based on Log-Polar Mapping and Phase Correlation, IEEE Transactions on Circuits and Systems for Video Technology, Special Issue on Authentication, Copyright Protection and Information Hiding, Vol., Issue, pp , August. Selected Journal Papers (quality evaluation based on digital watermarking) 7. Sha Wang, Dong Zheng, Jiying Zhao, Wa James Tam, and Filippo Speranza, An image quality evaluation method based on digital watermarking, IEEE Transactions on Circuits and Systems for Video Technology, Vol.7, No., pp.9-5, January 7.. Libin Cai, Ronghui Tu, Jiying Zhao, and Yongyi Mao, Speech quality evaluation: a new application of digital watermarking, IEEE Transactions on Instrumentation and Measurement, Vol. 5, No., pp. 5-55, February Sha Wang, Dong Zheng, Jiying Zhao, Wa James Tam, and Filippo Speranza, An Accurate Method for Image Quality Evaluation Using Digital Watermarking, IEICE Electronics Express (ELEX), Vol., No., pp.5-59, October 5.. Libin Cai and Jiying Zhao, Evaluation of speech quality using digital watermarking, IEICE Electronics Express (ELEX), Vol., No., pp. -5, October. Acknowledgement Graduate students Dong Zheng (RST invariant image watermarking, image quality evaluation) Yan Liu (RST invariant image watermarking) Ronghui Tu (audio watermarking, speech quality evaluation) Libin Cai (speech quality evaluation) Sha Wang (image/video quality evaluation) Colleagues Dr. Wa James Tam (Communications Research Centre Canada (CRC)) Dr. Filippo Speranza (CRC)

11 //7 Acknowledgement 井上智雄博士 筑波大学井上研究室 Research Center for Knowledge Communities 知的コミュニティ基盤研究センター