1 / 22

Joint Optimization of Data Hiding and Video Compression

Joint Optimization of Data Hiding and Video Compression. Jithendra K. Paruchuri & Sen-ching S. Cheung Department of Electrical and Computer Engineering Center for Visualization and Virtual Environments University of Kentucky, Lexington, KY 40507 ISCAS - May 21, 2008. Overview.

akando
Download Presentation

Joint Optimization of Data Hiding and Video Compression

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Joint Optimization of Data Hiding and Video Compression Jithendra K. Paruchuri & Sen-ching S. Cheung Department of Electrical and Computer Engineering Center for Visualization and Virtual Environments University of Kentucky, Lexington, KY 40507 ISCAS - May 21, 2008

  2. Overview • Motivation and Problem • Data Hiding Framework • Rate-Distortion Optimized Data Hiding • Results • Conclusion & Future Work www.vis.uky.edu | Dedicated to Research, Education and Industrial Outreach | 859.257.1257

  3. Signal Privacy • Smart video surveillance • Biometric theft • Mobile-media processing • RFID tracking

  4. Identity Obfuscation Original Black Box Pixelation/ Blurring Object Removal - Segmentation + Video In-painting [Venkatesh 06, 08]

  5. Problem with Obfuscation Police: Where were you 9am on Oct 1? A: I was in my office. Police: Do you have any proof? A: …… • Obfuscation destroys the authenticity • Original needed to legitimize the modification. • Accessed with proper authorization.

  6. Privacy Data Preservation • Separate Files or Meta-data • Cryptographic Scrambling [Boult05], [Dufaux06] • Data Hiding in DCT [Zhang05] • Works with any obfuscation techniques • High capability and low distortion • Fragile embedding • Eight-fold increase in output bit-rate!

  7. Parity Embedding Last decoded frame DCT Perceptual Mask • DCT Domain • Frequency, contrast and • luminance masking [Watson] Data Hiding with Compression DCT Entropy Coding Motion Compensation www.vis.uky.edu | Dedicated to Research, Education and Industrial Outreach | 859.257.1257

  8. Embedding & Perceptual Distortion • To embed x in a quantized DCT coeff. c(i,j,k) • Select coefficients to minimize distortion • Contrast masking • Frequency masking • Distortion

  9. Causes of Bit-rate Explosion • Disturbing the ‘favorable characteristics’ (zero blocks & long run-lengths) for entropy coding • Data embedding inserts noise into motion compensation loop • Proposed solution: Identify specific DCT coefficients for data hiding that minimizes both the output rate and distortion www.vis.uky.edu | Dedicated to Research, Education and Industrial Outreach | 859.257.1257

  10. R-D optimized Data Hiding Parity Embedding Last decoded frame Positions of the “optimal’ DCT coeff for embedding DCT R-D Optimization Perceptual Mask • DCT Domain • Frequency, contrast and • luminance masking [Watson] DCT Entropy Coding Parity Embedding Motion Compensation H.263 H.263 www.vis.uky.edu | Dedicated to Research, Education and Industrial Outreach | 859.257.1257

  11. Constrained Optimization Let Rk(Mk) and Dk(Mk) be the rate and distortion after hiding Mk bits into the k-th DCT block. δ is a user-defined weight Optimization Problem:

  12. Calculations of Rk(Mk) and Dk(Mk) • Two Issues: • Parity embedding depends on hidden data • Need to find optimal selection as well • “Worst-case” embedding on previous frame • Optimal selection • Dk(Mk) is additive (easy) • Rk(Mk) is not: Rk(N) and Rk(N+1) may use very different coefficients

  13. Dynamic Programming versus Greedy approach • Dynamic Programming • Greedy Approximation • Pick next coefficient to minimize cost • Fast implementation i i+1 i+1 i+2 i+2 i+3 i+3 Embedding K-th bit Embedding K+1-st bit

  14. Performance Comparison • For CIF, Greedy needs 26 second/frame vs. DP needs 22 minutes/frame

  15. Dual of the optimization • Lower bounded by unconstrained opt: • Search λ to meet constraint • Start at 2nd order approximation of Ck(Mk) • 3-5 seconds per one CIF frame

  16. Experiment 1: Hall Monitor

  17. Hall Monitor (QP=10) www.vis.uky.edu | Dedicated to Research, Education and Industrial Outreach | 859.257.1257

  18. Hall Monitor (QP=10) Original Distortion Optimized Weight = 0.5 Rate Optimized

  19. Surveillance Video

  20. Surveillance (QP=10) www.vis.uky.edu | Dedicated to Research, Education and Industrial Outreach | 859.257.1257

  21. Surveillance (QP=10) Original Rate Optimized Distortion Optimized Weight=0.5

  22. Conclusions • Privacy Data Preservation with Data Hiding • R-D Optimization Framework for Data Hiding • Current work • Reversible Data Embedding: ICIP 2008 • Privacy Data Management: ICIP 2008 • Incorporate temporal dimension in perceptual and rate model • Joint encryption and data hiding www.vis.uky.edu | Dedicated to Research, Education and Industrial Outreach | 859.257.1257

More Related