1 / 43

Multi-source Absolute Phase Estimation: A Multi-precision Approach Based on Graph Cuts

Multi-source Absolute Phase Estimation: A Multi-precision Approach Based on Graph Cuts. José M. Bioucas-Dias Instituto Superior Técnico Instituto de Telecomunicações Portugal. Graph Cuts and Related Discrete or Continuous Optimization Problems - IPAM 08. Phase Denoising (PD).

shirin
Download Presentation

Multi-source Absolute Phase Estimation: A Multi-precision Approach Based on Graph Cuts

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. Multi-source Absolute Phase Estimation:A Multi-precision Approach Based on Graph Cuts José M. Bioucas-Dias Instituto Superior Técnico Instituto de Telecomunicações Portugal Graph Cuts and Related Discrete or Continuous Optimization Problems - IPAM 08

  2. Phase Denoising (PD) Phase Unwrapping (PU) Estimation of Estimation of (wrapped phase) Absolute Phase Estimation

  3. Applications • Synthetic aperture radar/sonar • Magnetic resonance imaging • Doppler weather radar • Doppler echocardiography • Optical interferometry • Diffraction tomography

  4. Absolute Phase Estimation in InSAR (Interferometric SAR) InSAR Problem: Estimate 2- 1 from signals read by s1 and s2

  5. Mountainous terrain around Long’s Peak, Colorado Interferogram

  6. Differential Interferometry Height variation 7 mm/year -17 mm/year

  7. Magnetic Resonance Imaging - MRI Wrapped phase Intensity Interferomeric Phase • measure temperature • visualize veins in tissues • water-fat separation • mapthe principal magnetic field

  8. Outline • Forward problem (sensor model) • Absolute phase estimation: Bayesian formulation • Computing the MAP estimate via integer optimization • Multi-source absolute phase estimation • Phase unwrapping • Convex and non-convex priors • Unambiguous interval increasing • Phase unwrapping • Convex and non-convex priors

  9. Forward Problem: Sensor Model

  10. Simulated Interferograms Images of

  11. Data density: Prior (1st order MRF): clique set clique potential (pairwise interaction) non-convex convex Enforce smoothness Enforce piecewise smoothness (discontinuity preserving) Bayesian Approach

  12. posterior density • Phase unwrapping: Maximum a Posteriori Estimation Criterion

  13. Assume that Then PU ! summing over walks Phase Unwrapping: Path Following Methods Why isn’t PU a trivial problem? Discontinuities High phase rate Noise

  14. [Flynn, 97] (exact)! Sequence of positive cycles on a graph [Costantini, 98] (exact)! min-cost flow on a graph [Bioucas-Dias & Leitao, 01] (exact)! Sequence of positive cycles on a graph [Frey et al., 01] (approx)! Belief propagation on a 1st order MRF convex [Bioucas-Dias & Valadao, 05] (exact)! Sequence of K min cuts ( ) non-convex [Ghiglia, 96]! LPN0 (continuous relaxation) [Bioucas-Dias & G. Valadao, 05, 07] ! Sequence of min cuts ( ) Phase Unwrapping Algorithms

  15. while success == false then success == true PUMA (Phase Unwrapping MAx-flow) Finds a sequence of steepest descent binary images

  16. is submodular: each binary optimization has the complexity of a min cut • Related algorithms [Veksler, 99] (1-jump moves ) [Murota, 03] (steepest descent algorithm for L-convex functions) [Ishikawa, 03] (MRFs with convex priors) [Kolmogorov & Shioura, 05,07], [Darbon. 05](Include unary terms) [Ahuja, Hochbaum, Orlin, 03] (convex dual network flow problem) PUMA: Convex Priors • A local minimum is a global minimum • Takes at most K iterations

  17. Results ( )

  18. Results ( ) Convex priors does not preserve discontinuities

  19. PUMA: Non-convex priors Ex: Models discontinuities Models Gaussian noise Shortcomings: • Local minima is no more a global minima • Energy contains nonsubmodular terms (NP-hard) Tentative suboptimal solutions: • Majorization Minimization • Quadratic Pseudo Boolean Optimization (Probing [Boros et al., 2006], Improving [Rother et al., 2007] )

  20. Non-increasing property Majorizing nonsubmodular terms Majorization Minimization (MM) [Lange & Fessler, 95] [Rother et al., 05] ! similar approach for alpha expansion moves

  21. Interferogram no. of nonsubmodular terms iter us MM QOBOP QPBOI QPBOP MM QPBOI 1 590/0 2,5 e-2 590/0 590 326/0 1,0 e-2 2 326/0 410 263/0 1,0 e-2 263/0 271 3 154/0 6,0 e-3 154/0 179 4 123/0 4,0 e-3 123/0 141 5 94/0 4,0 e-2 6 94/0 117 88/0 2,5 e-3 88/0 91 7 57/15000 1,0e-3 57/15000 57 8 T 1 s 120 s 2 s Results

  22. Interferogram  MM QOBOP QPBOI Results

  23. Multi-jump version of PUMA Jumps 2 [1 2 3 4]

  24. PUMA + dyadic scaling then • Unary terms may be non-convex Compute using the algorithm [Darbon, 07] for 1st order submodular priors (complexity ) Absolute Phase (PU + Denoising) • Related algorithms: [Zalesky, 03], [Ishikawa, 03], [Ahuja, Hochbaum, Orlin, 04]

  25. Multi-source Absolute Phase Estimation

  26. Noise High phase rate Major degradation mechanism in PU and APE

  27. Use more than one observation with different frequencies Two sources We can infer • noise is an issue • unwrap phase images with range larger than Multi-source Absolute Phase Estimation

  28. Two sources

  29. Absolute phase estimation: • Phase v-unwrapping: Computing the MAP estimate

  30. Initialization: 1-unwrapp in the interval using total variation (TV) Optimization: Non-convex data term + TV Exact solution: Levelable functions [Darbon, 07], [Ahuja et al, 04], [Zalesky, 03], [Ishikawa, 03], (takes time) 2. Run PUMA in a multiscale fashion with the schedule: • scale v ! v-unwrapping] • scales ! denoising Proposed Algorithm

  31. for t=0:tmax success == false while success == false then success == true Absolute Phase (1-PU+v-PU + Denoising)

  32. High phase rate + noise

  33. Parabolic surface

  34. Future Directions • High order interactions • Denoise (first) + Unwrap • Local adaptive models (collaboration with Vladimir Katkovnik, Tampere University of Technology) • Huge images (ex: 10000£10000)

  35. Concluding Remarks • Addressed discontinuity preserving phase unwrapping and phase denoising methods based on integer optimization • Addressed multi-source absolute phase estimation • Introduced the concept of v-phase unwrapping • Introduced a new algorithm for multi-source absolute phase estimation based on integer optimization

  36. References • J. Dias and J. Leitao, “The ZM algorithm for interferometric image reconstruction in SAR/SAS”, IEEE Transactions on Image processing, vol. 11, no. 4, pp. 408-422, 2002. • J. Bioucas-Dias and G. Valadao, “Phase unwrapping via graph cuts", IEEE Transactions on Image processing, vol. 16, no. 3, pp. 698-709, 2007. • V. Kolmogorov and A. Shioura, “New algorithms for the dual of the convex cost network flow problem with applications to computer vision", Technical Report, June, 2007 • J. Darbon, Composants logiciels et algorithmes de minimisation exacte d’energies dedies au traitement des images, PhD thesis, Ecole Nationale Superieure des Telecommunications, 2005. • Y. Boykov and Vladimir Kolmogorov, An Experimental Comparison of Min-Cut/Max-Flow, IEEE Transactions on Pattern Analysis and Machine Intelligence, September 2004.

  37. References • C. Rother, V. Kolmogorov, V. Lempitsky, and M. Szummer, “Optimizing binary MRFs via extended roof duality”, in IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2007. • E. Boros, P. L. Hammer, and G. Tavares. Preprocessing of unconstrained quadratic binary optimization. Technical Report RRR 10-2006, RUTCOR, Apr. 2006. • J. Darbon and M. Sigelle, “Image restoration with discrete constrained total variation Part II: Levelable functions, convex and non-convex cases”, Journal of Mathematical Imaging and Vision. Vol. 26 no. 3, pp. 277-291, December 2006. • B. Zalesky, “Efficient determination of Gibbs estimators with submodular energy functions”, arXiv:math/0304041v1 [math.OC] 3 Apr 2003.

  38. Acknowledgements Gonçalo Valadão Yuri Boykov Vladimir Kolmogorov

More Related