Chen Bar Ioannis Gkioulekas and Anat Levin Technion Israel CMU USA Speckle image Coherent scattering and memory effect 2 Tissue Microscope objective Speckle Image Scattering Layer ID: 997779
Download Presentation The PPT/PDF document "Rendering Near-Field Speckle Statistics ..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
1. Rendering Near-Field Speckle Statistics in Scattering MediaChen Bar, Ioannis Gkioulekas and Anat LevinTechnion, IsraelCMU, USA
2. Speckle imageCoherent scattering and memory effect 2TissueMicroscope objective
3. Speckle ImageScattering LayerHidden ObjectSpeckle imaging applications Seeing through scattering materials [Katz et al. 2014]Applications and related work3
4. 4Speckle imaging applications For what types of materials does this work?What kinds of targets can it reconstruct?Speckle ImageScattering LayerReconstructionSeeing through scattering materials [Katz et al. 2014]Hidden Object
5. Spatial light modulatorSpeckle imaging applications 5Adaptive optics focusing through scattering material [Judkewitz et al. 2014]Can we focus at nearby points by reusing the same adaptive pattern?For what ranges is this possible?
6. Rendering speckles, how?Solve wave optics equationsMonte-Carlo approachBar et al. 2019ImpracticalPhysically accurateOrders of magnitude faster6
7. Far field v.s. near field Point sourceImaged point sourceLensView pointWide collimated illuminationDirectional viewTissue imaging applications are in the near fieldpoint source inside the materialLensImaged view pointNear fieldFar field7our previous algorithm assumes far-field
8. Visualizing near-field covarianceIlluminations8Viewing objective
9. IlluminationsVisualizing near-field covariance 9Viewing objective
10. Intensity rendering 10Near fieldFar field
11. Covariance rendering 11Near fieldFar fieldMain challenge: efficiently estimating the total throughput through the apertures
12. Near-field rendering? Potential solution 1 Bad approach #1: compute multiple tabulated values of far-field covariance, then apply linear transformation to compute the near field covarianceLots of far-field directions to render. Huge computation and memory requirements 12
13. Bad approach #1: compute multiple tabulated values of far-field covariance, then apply linear transformation to compute the near field covarianceLots of far-field directions to render. Huge computation and memory requirements Near-field rendering? Potential solution 1Tabulation approachOur approach15 seconds100 minutes25 minutesFewer samples -> aliasing13
14. Near-field rendering? Potential solution 2 14Computation involves integrals over the 4 apertures Bad approach #2: Monte Carlo sampling Sample directionscomplex-valued integrals -> High variance
15. Bad approach #2: Monte Carlo sampling complex-valued integrals -> High varianceSample directionsNear-field rendering? Potential solution 2Monte-Carlo aperture integration approachOur approach15 seconds15 seconds150 seconds15
16. Near-field rendering: our approach Closed-form approximation 16
17. Near-field rendering: contributionsClosed-form spherical integration using von Mises-Fisher (vMF) representation of aperture functions and phase functions Binary aperturevMF approximationTissue phase functionvMF fit17hemisphere projection
18. Closed-form spherical integration using von Mises-Fisher (vMF) representation of aperture functions and phase functions Binary aperturevMF approximationTissue phase functionvMF fithemisphere projectionNear-field rendering: contributions18See paper for detailsComplex von Mises-Fisher function definition Complex von Mises-Fisher function integrationComplex von Mises-Fisher function convolution Efficient algorithmNegligible bias
19. Near-field rendering: contributionsClosed-form spherical integration using von Mises-Fisher (vMF) representation of aperture functions and phase functions Efficient importance sampling strategy Where should we sample path starting points?Imaged point sourceZero contributionHigh contribution19
20. Convergence time comparisonOur near fieldTabulation approachMonte-Carlo aperture integration approach 15 seconds1.6 hours14 hoursTime to 20
21. 21Equal-time comparison1 seconds5 seconds15 seconds Difference scales with the size of the target volume Our near fieldTabulation approachMonte-Carlo aperture integration approach
22. 22Comparison with lab measurements Shift Tilt Analytical model[Osnabrugge et al. ]Lab measurement[Osnabrugge et al. ]Monte Carlo simulation[Ours]
23. 23Focusing through turbid media: range predictions Spatial light modulatorAdaptive optics focusing through scattering material [Judkewitz et al. 2014]For what ranges is this possible?
24. Spatial light modulatorAdaptive optics focusing through scattering material [Judkewitz et al. 2014]For what ranges is this possible? 24Focusing through turbid media: range predictions OD 3Analytical ModelOur simulationO.D. 1O.D. 3
25. 25SummaryCritical tissue imaging applications require near-field simulations rather than far field.Efficient adaptation of far-field rendering algorithms to the near-field based on closed-form von Mises-Fisher integration.Orders of magnitude improvement compared to alterative techniques.Evaluation on focusing through scattering applications.Speckle ImageScattering LayerReconstructionHidden Object
26. 26Thank you!Open source code is available onlineRendering Near-Field Speckle Statistics in Scattering Media TissueSensorThis work was supported by:https://github.com/chabner/gaussianBeam-field