S S S A2009 Simulation Study Of Segmentation
Upcoming SlideShare
Loading in...5
×
 

S S S A2009 Simulation Study Of Segmentation

on

  • 571 views

 

Statistics

Views

Total Views
571
Views on SlideShare
570
Embed Views
1

Actions

Likes
0
Downloads
3
Comments
0

1 Embed 1

http://www.slideshare.net 1

Accessibility

Categories

Upload Details

Uploaded via as Microsoft PowerPoint

Usage Rights

© All Rights Reserved

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Processing…
Post Comment
Edit your comment

S S S A2009 Simulation Study Of Segmentation S S S A2009 Simulation Study Of Segmentation Presentation Transcript

  • A Simulation Study of Segmentation Methods on the Soil Aggregate Microtomographic Images Wei Wang, Alexandra N. Kravchenko, Kateryna Ananyeva, Alvin J. M. Smucker, C.Y. Lim and Mark L. Rivers Department of Crop & Soil Sciences, Department of Statistics & Probability, MSU Advanced Photon Source, Argonne National Laboratory
  • Computed microtomographic images (CMT)
  • Motivation
    • Segmentation results will affect pore network analysis, (e.g. pore connectivity, tortuosity, medial axis … ) and therefore influence flow simulation (e.g. Lattice Boltzmann modeling ), pore-scale biological activity modeling.
    • Accurate pore/solid classification is very important to understand pore structures in the intra-aggregate spaces.
  • Difficulties in processing CMT images
    • Artifacts : partial volume effect (finite resolution effect), beam hardening, ring artifacts …
    • Complex composition of soil matrix (large range of grey-scale values)
  • Difficulties in processing CMT images
    • Lack of ground-truth information to assess pore/solid classification accuracy
    • Identify criteria to select optimal segmentation method for soil aggregate images
  • Objectives
    • To evaluate criteria for selecting the optimal segmentation method for soil pore characterization
    • To compare performance of several commonly used segmentation methods in soil aggregate images with different porosities
  • Simulation approach
    • In order to overcome the absence of ground-truth information we proposed a simulation approach including:
    • ● Simulate partial volume effect in the pore space
    • ● Simulate different solid material
    • ● Simulate random noise
  • Simulate partial volume effect
    • Generate binary soil images at the scanned resolution;
    scanned pixel size Ground-truth image 1 mm
  • Simulate partial volume effect
    • Generate 3 layers of pores at smaller scales;
    1/2 scanned pixel size 1 mm 1/8 scanned pixel size 1 mm 1/4 scanned pixel size 1 mm
  • Simulate partial volume effect
    • Combine all the layers of pores
    1 mm
  • Simulate solid space and noise
    • Solid space simulation was done for all the “ white ” pixels using spatial simulation of LU decomposition technique
    • Gaussian random noise was added to the whole image
  • Grey scale image simulation Ground truth image Simulation in the pore space Simulation in the solid space + noise simulation Original image from the scan
  • Different porosity cases (1) Low Medium High High + flow pattern Porosity = 4.8% Porosity = 7.8% Porosity = 16.5% Porosity = 22.8%
  • Different porosity cases (2) Low Medium High High + flow pattern Porosity = 3.6% Porosity = 8.3% Porosity = 15.8% Porosity = 28.5%
  • Existing segmentation methods
    • More than 40 different segmentation methods (Sezgin et al., 2004 )
    • They mainly can be classified into several categories:
    • ● Manual thresholding
    • ● Global thresholding methods
    • ● Local adaptive methods
  • Segmentation methods
    • Global thresholding :
    • ● Entropy method: Renyi ’ s entropy (Sahoo et al., 1997)
    • ● Iterative method: Riddler et al., 1978
    • ● Otsu method: maximize between-class variance (Otsu, 1979)
    • Local adaptive method:
    • ● Indicator kriging ( IK ) method: Oh and Lindquist, 1999
  • IK method
    • Two steps : thresholding, kriging
    • Thresholding step: the thresholds are determined by fitting mixed Gaussian distributions to pore and solid spaces using Expectation-Maximization algorithm (Dempster et al., 1977 ).
    Black White T 1 T 2 Solid Pore kriging step ?
  • Segmentation performance criterion
    • Misclassification Error (ME): 0<ME<1 (ground-truth image required)
    • where P and S are the number of common pore or solid pixels in both ground-truth and segmented images.
  • Segmentation performance criterion
    • Region non-uniformity measure (NU): 0<NU<1 (ground-truth image not required)
    • Where P and T are the numbers of pore and total number of pixels in the segmented image, and are the variance of grey-scale values in the pore space and total variance in the simulated grayscale image.
    Whether NU can be used as a criterion for soil ?
  • How good is NU for soil ?
    • Pore morphological characteristics:
    • Porosity
    • Number of connected pores
    • Number of pore boundary pixels
    • Number of pore skeleton pixels
  • Results (Low porosity) Ground truth image IK Entropy Iterative Otsu Distinct segmentation error
  • Results (Medium porosity) Ground truth image IK Entropy Iterative Otsu
  • Results (High porosity) Ground truth image IK Entropy Iterative Otsu
  • Results (High+flow pattern) Ground truth image IK Entropy Iterative Otsu
  • Comparisons of segmentation methods using ME and NU Overall ranking by ME : IK > Entropy > Iterative > Otsu Overall ranking by NU : IK > Otsu > Iterative > Entropy Indicator Kriging is the best! Indicator Kriging is the best! IK Iter Otsu Entropy Entropy IK Otsu Iter
  • How good is NU for preserving pore characteristics ?
    • * Relative error = ( the pore characteristic value from the segmented image - the pore characteristic ground-truth value)/ the ground-truth value
  • Summary
    • Soil aggregate CMT images were generated from the pore/solid binary image by simulating partial volume effect, different solid material and background noise.
    • No single method preserved pore characteristics in all cases. However, Indicator Kriging method yielded segmented images most similar to the ground-truth images in the majority of cases studied.
  • Summary
    • We recommend using NU as a criterion for choosing best segmentation approaches.
    • Segmentation assessment using NU provides acceptable representation of pore characteristics in the segmented images.
    • USDA-CSREES National Research Initiative:
    • Project 2008-35102-04567
    • NSF LTER Program at Kellogg Biological Station and the Michigan Agricultural Experiment Station
    • Advanced Photon Source, Argonne National Lab
    Acknowledgement
  • Thanks for your attention!
  • References
    • M. Sezgin, B. Sankur. 2004. Survey over image thresholding techniques and quantitative performance evaluation. Journal of Electronic Imaging 13(1), 146 – 165
    • P. Sahoo, C. Wilkins and J. Yeager. 1997. Threshold selection using Renyi ’ s entropy. Pattern Recognition, Vol.1, No.1, 71-84
    • W. Oh, B. Lindquist. 1999. Image thesholding by indicator kriging. IEEE Transactions on Pattern Analysis and Machine Intelligence 21: 590-602.
    • T. W. Ridler and S. Calvard, ‘‘ Picture thresholding using an iterative selection method, ’’ IEEE Trans. Syst. Man Cybern. SMC-8, 630 – 632 ~1978.
    • N. Otsu (1979). &quot;A threshold selection method from gray-level histograms&quot;. IEEE Trans. Sys., Man., Cyber. 9: 62 – 66
    • Dempster, A.P., Laird, N.M. and Rubin, D.B., 1977. Maximum likelihood from in- complete data via the em algorithm. Journal of the Royal Statistical Society: Series B, 39(1): 1-38.