Sumanth Jagannathan

1
Distributed Compression of Lightfields

• Sumanth Jagannathan
• and
• Primoz Skraba

2
Overview

• Motivation
• Distributed Compression of Lightfields
• Rate Control
• Wavelets
• Conclusion

3
Lightfields

• Lightfield data of one object from many
  views
• Large amount of correlated data
• Distributed compression
• Lower bandwidth requirement
• Lower image sensor complexity
• Wyner-Ziv coding
• Thought of as a Slepian-Wolf coder followed by a
  quantizer
• Done in practice with turbo coder
• Use Key views to generate side information at the
  decoder
• Assume that we have perfect geometry information

4
Gains of Distributed Compression
5
Rate Control

• Problem definition
• Obtain desired quality given the total rate of
  key views and Wyner-Ziv views
• Many degrees of freedom
• Our approach
• Fix quality of key views
• Choose quantization step
  size of W-Z views
• Implementation
• Effect of bit rate on side information quality

6
Rate Control Results
7
Rate Control Results
8
Wavelet Transform Coder

• Wyner-Ziv coding in the transform domain
• Low frequency sub-bands should be well
  correlated, resulting in low Wyner-Ziv rates
• Higher frequency sub-bands are quantized with
  lower resolution (compression gain)

9
Conclusion

• Gains at low bit rates using Wyner-Ziv coding
• For rate control
• Quality of key views determines Wyner-Ziv
  quantization step size
• Number of key views is very important
• Optimal number of key frames exists which will
  maximize PSNR for a total given rate
• For Wavelets
• Simulations give poor performance
• Entropy based calculations indicate some gains
• Simulation model and turbo encoder/decoder needs
  to be re-examined

10
Acknowledgements

• We would like to thank Prof. Girod, Xiaoqing,
  Anne, David, Prashant, and Shantanu for their
  helpful discussions.

11
Bit Allocation
12
(No Transcript)
13
Bitplane Extraction