growing least squares for the analysis of manifolds in
play

Growing Least Squares for the Analysis of Manifolds in Scale-Space - PowerPoint PPT Presentation

Growing Least Squares for the Analysis of Manifolds in Scale-Space Nicolas Mellado , Gal Guennebaud, Pascal Barla, Patrick Reuter, Christophe Schlick Inria - Univ. Bordeaux - IOGS - CNRS nicolas.mellado@inria.fr Context Shape


  1. Growing Least Squares for the Analysis of Manifolds in Scale-Space Nicolas Mellado † , Gaël Guennebaud, Pascal Barla, Patrick Reuter, Christophe Schlick Inria - Univ. Bordeaux - IOGS - CNRS † nicolas.mellado@inria.fr

  2. Context  Shape matching [ HFGHP06 ,IT11]  Focus on geometric properties  Describe shape features  Find their pertinent scales  Requires a multi-scale analysis 2

  3. Problem statement  Find pertinent structures in scale-space  Process  Ignore noise  Extract points  Detect similarity Scale Arc-length 3

  4. Previous work - 1/3  Intrinsic characterization of the shape:  Heat diffusion (HKS) [ SOG09 ,BK10, ZRH11] 4

  5. Previous work - 1/3  Intrinsic characterization of the shape:  Heat diffusion (HKS) [ SOG09 ,BK10, ZRH11] Medium to global scales We want: • Fine to medium • Pertinent scale extraction 5

  6. Previous work - 2/3  Multi-scale geometric descriptor: curvature  Mean Curvatures estimated via smoothing [ ZBVH09,MFK ∗ 10 ]  Covariance Analysis [ PKG2003 , LG2005, YLHP06] 6 Growing Least Squares

  7. Previous work - 2/3  Multi-scale geometric descriptor: curvature  Mean Curvatures estimated via smoothing [ ZBVH09,MFK ∗ 10 ] Curvature is not enough…  Covariance Analysis [ PKG2003 , LG2005, YLHP06] 7 Growing Least Squares

  8. Previous work - 3/3  Scale-space analysis [ ZBVH09,MFK ∗ 10 ,DK11]  Curvature extrema extraction 8 Growing Least Squares

  9. Previous work - 3/3  Scale-space analysis [ ZBVH09,MFK ∗ 10 ,DK11]  Curvature extrema extraction Not exhaustive pertinence detection 9 Growing Least Squares

  10. Our approach: Growing Least Squares  Regression method: fit hyper-sphere  Multi-scale  Support huge point-sets  Characterization by 2 nd order proxy  Analytic detection of pertinent scales  Dense description  Bonus  2D curves  3D surfaces 10

  11. Our approach: Growing Least Squares 11

  12. Pipeline Geometric descriptor based on 2 nd -order regression, 1. Continuous & analytic geometric variation in scale. 2. 2 1 12

  13. Geometric Descriptor – 1/3  Algebraic hyper-sphere fitting procedure [GG2007] w 1 τ κ η {x ; S u (x) = 0} 13

  14. Geometric Descriptor – 1/3  Algebraic hyper-sphere fitting procedure [GG2007] w  Normalization 1 τ κ η {x ; S u (x) = 0} 14

  15. Geometric Descriptor – 1/3  Algebraic hyper-sphere fitting procedure [GG2007] w  Normalization 1 τ κ η  Re-parametrization {x ; S u (x) = 0} û = [ τ /t η t κ ] T 15

  16. Geometric Descriptor – 1/3  Algebraic hyper-sphere fitting procedure [GG2007] w  Normalization 1 τ κ η  Re-parametrization {x ; S u (x) = 0}  Scale Invariance û = [ τ /t η t κ ] T 16

  17. Geometric Descriptor – 2/3  Impact of τ  Impact of η 17

  18. Geometric Descriptor – 3/3  An example in 3D 18

  19. Geometric Descriptor – 3/3  An example in 3D 19

  20. Geometric Variation  Given our descriptor [ τ η κ ] T d τ d η d κ  Variation is related to [ ] T dt dt dt  Geometric Variation 20

  21. Summary  Geometric Descriptor  Geometric Variation 21

  22. Results/Applications – 1/3  Descriptor is robust to noise  Can guide surface reconstruction Gargoyle - 250k points 5% random noise 20 scales, 6.6 s . 22

  23. Results/Applications – 1/3  Descriptor is robust to noise  Can guide surface reconstruction 23

  24. Results/Applications – 2/3  Continuous detection of the feature area 24

  25. Results/Applications – 2/3  Continuous detection of the feature area 2d curve- 5k points Armadillo - 173k points 1000 scales, 3 s . 20 scales, 6.3 s . 25

  26. Results/Applications – 3/3  Similarity detection,  Multi-scale profile 26

  27. Results/Applications – 3/3  Similarity detection,  Multi-scale profile  Specifics scale range Torus- 500k points 20 scales, 42 s . 27

  28. Discussion – 1/2  Main limitation  isotropic,  Analytic fitness measure  Squared Pratt Norm [GG2007]  φ = || u l || 2 - 4u c u q 28

  29. Discussion – 2/2  Use multi-scale profile to characterize shape 29

  30. Conclusions  Contributions  Stable second order descriptor for 2D/3D manifolds  Continuous in scale and space from point-set input  Continuous analysis to detect pertinent scales  Future work:  Use non-linear kernel to deal w/ additional attributes  Analyze spatial variations to characterize anisotropy 30

  31. Growing Least Squares for the Analysis of Manifolds in Scale-Space Thank you Nicolas Mellado Gaël Guennebaud Pascal Barla Patrick Reuter Christophe Schlick www.labri.fr/perso/mellado Inria - Univ. Bordeaux - IOGS – CNRS http://manao.inria.fr Project ANR SeARCH European Consortium v-must.net 31

Recommend


More recommend