material images: from raw data to physical measurements Chloe - - PowerPoint PPT Presentation

material images from raw data to
SMART_READER_LITE
LIVE PREVIEW

material images: from raw data to physical measurements Chloe - - PowerPoint PPT Presentation

Segmentation of 3-D glass material images: from raw data to physical measurements Chloe Brillatz Surface du Verre et Interfaces 15/11/2018 The motivations 128 microns barium borosilicate glass SiO 2 + BaO + B 2 O 3 2 The motivations Thin


slide-1
SLIDE 1

Chloe Brillatz Surface du Verre et Interfaces 15/11/2018

Segmentation of 3-D glass material images: from raw data to physical measurements

slide-2
SLIDE 2

2

The motivations

128 microns

barium borosilicate glass SiO2 + BaO + B2O3

slide-3
SLIDE 3

3

SVI – CONFIDENTIAL – Disclosure or reproduction without prior written permission of Saint-Gobain Recherche is prohibited

The motivations

Main glass Thin layer Nano-texture Super hydrophobic glass

slide-4
SLIDE 4

4

From raw glass to physical measurements

slide-5
SLIDE 5

5

SVI – CONFIDENTIAL – Disclosure or reproduction without prior written permission of Saint-Gobain Recherche is prohibited

Data acquisition

3D imaging by X-ray tomography at the Grenoble Synchrotron (European Synchrotron Radiation Facility):

 Non-destructive method + evolution in time  900°C  Nanometric resolution ( ESRF Beam 16b)

slide-6
SLIDE 6

SVI – CONFIDENTIAL – Disclosure or reproduction without prior written permission of Saint-Gobain Recherche is prohibited

128 microns 108 microns

slice view

Full glass sample

Data: horizontal view

Movie

slide-7
SLIDE 7

7

SVI – CONFIDENTIAL – Disclosure or reproduction without prior written permission of Saint-Gobain Recherche is prohibited

The goal

slide-8
SLIDE 8

8

SVI – CONFIDENTIAL – Disclosure or reproduction without prior written permission of Saint-Gobain Recherche is prohibited

Raw Image

slide-9
SLIDE 9

9

SVI – CONFIDENTIAL – Disclosure or reproduction without prior written permission of Saint-Gobain Recherche is prohibited

After local filter

ndimage.median_filter from scipy

  • Computes the median
  • f the voxel values in a given

window, the central voxel is replaced with this median value

After Before

slide-10
SLIDE 10

10

SVI – CONFIDENTIAL – Disclosure or reproduction without prior written permission of Saint-Gobain Recherche is prohibited

After segmentation – 2D view – unsuccessful test

threshold_otsu from skimage.filters

  • Reduces a gray level image

to a binary image

  • globally calculates the

threshold

slide-11
SLIDE 11

11

SVI – CONFIDENTIAL – Disclosure or reproduction without prior written permission of Saint-Gobain Recherche is prohibited

After segmentation – 2D view

rank.otsu from skimage.filters

  • Reduces a gray level image

to a binary image

  • locally calculate the

threshold

slide-12
SLIDE 12

12

SVI – CONFIDENTIAL – Disclosure or reproduction without prior written permission of Saint-Gobain Recherche is prohibited

Raw image

slide-13
SLIDE 13

13

SVI – CONFIDENTIAL – Disclosure or reproduction without prior written permission of Saint-Gobain Recherche is prohibited

The goal

Movie

mayavi 3D visualization mlab.pipeline.contour_grid_plane grid_plane.sync_trait

slide-14
SLIDE 14

14

SVI – CONFIDENTIAL – Disclosure or reproduction without prior written permission of Saint-Gobain Recherche is prohibited

Some numbers

x 80

Time: 0 s Time: 4 hours

…………..

1.8 billions voxels

7 GB Total 560 GB Parallel from joblib 50 chunks

  • verlap = 10 slices

30 cores Total running time for the segmentation: 17 hours

slide-15
SLIDE 15

15

SVI – CONFIDENTIAL – Disclosure or reproduction without prior written permission of Saint-Gobain Recherche is prohibited

After segmentation – 3D view

marching_cubes_lewiner from skimage.measure

  • finds surfaces in 3d

volumetric data mlab.triangular_mesh from mayavi

slide-16
SLIDE 16

16

SVI – CONFIDENTIAL – Disclosure or reproduction without prior written permission of Saint-Gobain Recherche is prohibited

After segmentation – 3D view

Movie

marching_cubes_lewiner from skimage.measure

  • finds surfaces in 3d

volumetric data mlab.triangular_mesh from mayavi

slide-17
SLIDE 17

17

SVI – CONFIDENTIAL – Disclosure or reproduction without prior written permission of Saint-Gobain Recherche is prohibited

Labelling

label from scipy morphology.remove_small_objects from skimage

slide-18
SLIDE 18

18

SVI – CONFIDENTIAL – Disclosure or reproduction without prior written permission of Saint-Gobain Recherche is prohibited

Data analysis

slide-19
SLIDE 19

19

SVI – CONFIDENTIAL – Disclosure or reproduction without prior written permission of Saint-Gobain Recherche is prohibited

After segmentation – 4D view

Movie

TimeVisualizer - Mayavi

  • A GUI for 3-D visualization

and exploration of a time series of 3-D images

slide-20
SLIDE 20

20

SVI – CONFIDENTIAL – Disclosure or reproduction without prior written permission of Saint-Gobain Recherche is prohibited

What to take home

start simple compromise between quality and time Python simple and powerful tools

1 TB

slide-21
SLIDE 21

21

SVI – CONFIDENTIAL – Disclosure or reproduction without prior written permission of Saint-Gobain Recherche is prohibited

Contact

I’d be happy to share more details, the codes and ideas: Please contact me: chloe.brillatz@gmail.com