for compressed imaging
play

for Compressed Imaging Chunli Guo, Mike E. Davies Institute for - PowerPoint PPT Presentation

A Sample Distortion Analysis for Compressed Imaging Chunli Guo, Mike E. Davies Institute for Digital Communications University of Edinburgh, UK 1 Talk Outline Introduction Compressed Sensing: from Sparse to Compressible, from


  1. A Sample Distortion Analysis for Compressed Imaging Chunli Guo, Mike E. Davies Institute for Digital Communications University of Edinburgh, UK 1

  2. Talk Outline • Introduction • Compressed Sensing: from Sparse to Compressible, from Deterministic to Stochastic • Sample Distortion (SD) framework • definition, examples, lower bounds and convexity • Multi-resolution CS • Wavelet Statistical Image Model • SD and Optimal Bandwise Sampling • Oracle Bounds • Sample Allocation with tree structure • Natural Image Examples • Conclusions 2

  3. Sparsity and Compressed Sensing 3

  4. Compressed sensing Compressed Sensing assumes a Set of signals of sparse/compressible set of signals interest Uses random projections for observation matrices Signal reconstruction by a nonlinear mapping. Compressed sensing provides nonlinear approximation practical algorithms with guaranteed random projection (reconstruction) performance e.g. L1 min., OMP, (observation) CoSaMP, IHT. Closely linked with theory of n- widths

  5. The L1 solution guarantee

  6. Compressible vectors (deterministic)

  7. Compressible Distributions

  8. A Sample-Distortion framework for CS

  9. Sample Distortion Framework [Guo & D. 2011/2012]

  10. Sample Distortion Framework 10

  11. SD Functions for 2-state GSM   ( ) 0.38 (0,1.198) 0.62 (0,0.0044) p x N N   ( ) 0.38 (0,1.198) 0.62 (0,0.0044) p x N N Convexity implies achievable (by zeroing) BAMP SD fun BAMP SD fun MBB EBB 11

  12. SD Lower Bounds 12

  13. Example: Generalized Gaussian Gaussian EBB Laplace EBB GGD, α =0.4 EBB Wavelet coefficients of natural images are often modelled as GGD with α ≈ 0.4 -1.0

  14. SD Lower Bounds 14

  15. Convexity of D( δ ) Theorem: The SD function, D( δ ), is convex Convex hull achievable by         , combinations of and , 1 1 1 1     D , 1 D 2 2 3     , 2 2 D 2    1 3 2 15

  16. Gaussian encoders are not optimal! Folk theorem - Gaussian encoders are optimal. False ! If Gaussian-specific SD function is not convex we can do better 16

  17. SD Functions for 2-state GSM   ( ) 0.38 (0,1.198) 0.62 (0,0.0044) p x N N Convexity implies achievable (by zeroing) BAMP SD fun MBB EBB 17

  18. Multi-resolution Compressive Imaging

  19. Statistical image model 19

  20. Test Images

  21. Image model example cameraman

  22. Image model example GGD representation with Haar wavelets... Estimated shape parameters for each level cameraman

  23. Image model example GGD wavelet representation Estimation of the variance for each level e.g. cameraman

  24. Bandwise Compressive Imaging

  25. Bandwise sampling

  26. Optimal Bandwise sample allocation

  27. Bandwise Sampling Convexified MMSE AMP distortion reduction function (band 1 for cameraman image model) distortion reduction distortion 27

  28. Bandwise Sample Allocation DR fun for cameraman image 28

  29. Bandwise CS sample allocation Sample allocation (% of full sampling) per band for m = 170, 600, 2000 and 10000 measurements. There are typically no more than 2-3 partially sampled bands

  30. Bandwise CS Performance e.g. cameraman 30

  31. adding Tree Structure

  32. Incorporating Tree Structure We can add tree-based priors on coefficients and decode using Turbo AMP scheme [Som, Schniter 2012]: This calculates marginal probabilities for hidden states and incorporate into MMSE AMP 32

  33. Bandwise CS Sample Allocation Sample allocation (% of full sampling) per band for δ= 10%, 15.26%, 25% and 30% m=1000 m=6554 SA for cvx SD fun Empirical best SA m=19661 with tree info m=16384 SA for SD fun with oracle tree info 33

  34. Bandwise CS Performance e.g. cameraman 34

  35. Reconstructed I Image reconstructions from 10000 measurements (15%) 35

  36. SA for General Image Statistics 36

  37. Open questions • How to derive sample allocations for more sophisticated models? – analysis representations, tree structured model, etc. • How to allocate samples within constrained sampling schemes (e.g. partial Fourier)? 37

  38. References • R. Gribonval, V. Cevher, and M. E. Davies. Compressible Distributions for High-dimensional Statistics. Preprint, 2011, available at arXiv:1102.1249v2. • M. E. Davies and C. Guo, Sample-Distortion Functions for Compressed Sensing , 49th Allerton Conf. on Communication, Control, and Computing, 2011. • C. Guo and M. E. Davies , Sample-Distortion for Compressed Imaging , on IEEE TSP, 2013.

  39. Thank You 39

Recommend


More recommend