The 2012 World Congress on Advances in Civil, Environmental, and Materials Research (ACEM’ 12)Seoul, Korea, August 26-30, 2012

Image processing techniques relevant to geomaterials

Dong Hun Kang1), Tae Sup Yun2), and Kwang Yeom Kim3)

1) Department of Civil and Environmental Engineering, Yonsei University, Seoul 120-749, Korea

2) Department of Civil and Environmental Engineering, Yonsei University, Seoul 120-749, Korea,

3) Korea Institute of Construction Technology, Goyang 411-712, Korea

ABSTRACT

X-ray CT images provide the spatial configuration of pore space in soils and internal void distribution of porous geomaterials. The pore images qualitatively visualize the random and heterogeneous pore structure with the lack of quantitative description of pore size distribution. Moreover, the CT images inherently include the unavoidable noises such as beam hardening and ring artifacts. This study presents the image processing techniques applicable to CT images for geomaterials. Pore structures are quantified by the pore chambers and channel with the aid of Delaunay tessellation, Euclidean distance transformation, pore mergence, and A-star algorithm, which results in the evaluation of pore connectivity and pore size distribution. Noises are reduced with the calibration of pixel consistency, coordinate transformation, and Fourier transformation. The image segmentation is enabled by binarization. Examples of granular soils and construction materials are presented to highlight the applicability and implication to enhance the quality of CT images under analysis. 1. INTRODUCTION

The 3D X-ray computed tomography naturally provides the qualitative information of internal microstructure of target geomaterials, which becomes fundamental information of geometry for quantifying heterogeneously and randomly configured geometry(Kikuch et.al., 2010). The geometrical configurations under analysis include the heterogeneous and irregularly shaped particle shape and its network, interconnected pore structures, and the existence of discontinuity such as shear band and fractures under loading, mostly as qualitative characters. One of the most common examinations is the

1) Graduate Student 2) Assistant Professor (corresponding author) 3) Senior Researcher

segmenpropertiespacing to the drange ofintroducapplicab

2. NOIS 2.1 B WhileincreaseFig.1(a) for an expixel valto the bthresholdependieither un (a)

Fig. 1 (aof por

We tranradial nowhereasplotted ifitted witwith resoriginallyinconsis

tation of des related factors, et

defect of hf software

ces the noble to pract

ES AND E

Beam Harde the X-rayes so that t

shows thaxample of lues shoulbeam hardd deteriorang on thenderestima

a) Beam hare space ei

sform the oise to stris y-axis dein Fig 2(b)th 4th polyspect to thy located

stency decr

different pto spatial ctc). Howevhardware aand hardw

oises commtice.

ENHANCE

dening Effey energy pthe pixel vaat the peripcement-bad become dening effates createe thresholdated or ove

(b)

ardening efither under

image in Cip one as enotes the) and the pynomial funhe referen

at the imreases.

hases basconfigurativer, it is unand the inware calibrmonly prev

EMENT

ect penetrates alues at thphery of a ased mate0 while hi

ect (e.g., es the imad value. Inerestimated

ffect showrestimated

Cartesian toshown in

e radius ofprofile is anction by lece value Image cen

sed on thon of phasavoidable completen

rations canvailing an

and passe center te cylindrica

erial. Depeigher onesoften call

age qualitynstead of d in radial

(c)

n in cemen or overes

o that in PFig 2(a) wf original ialike to theeast squarIc. This prnter to th

e thresholses (e.g., pto exclude

ness of invn improve td the enh

ses througend to be lol specimen

ending on ts define theed cuppin. Figs 1b-dclear segmdirection.

nt-based mtimated by

Polar Cartewhere x-axi

mage. Thee cupping. e method. rocedure se higher

ld value wporosity, poe noises froversion algthe image hancement

h the objeower(Barren is lighterthe threshoe solid phag effect),

d presents mentation,

(d

material. (by different t

sian spaces indicatese average To calibraThen, pixe

shifts the level so

which provore size disom CT imagorithm altquality. Th

t technique

ect, the abett and Kear in radial dold on valuase. Howea single

the binary pores ten

d)

b)-(d). Binathreshold v

e, which ms the radia values alate it, the el values alower pixethat the

vides the stribution, ages due though a his paper e readily

bsorption at 2004). direction, ue, lower ever, due value of

y images nd to be

ry image values.

makes the al angles long θ is profile is

are offset el values intensity

(a)

(b) (c)

Fig. 2 (a). Polar transformed image of specimen. (b) Mean pixel value profile of original image. (c) Mean pixel value profile of corrected image.

2.2 Ring Artifact

The defect on the detector and the tomographic inversion are attributed to existence of the radial and periodic noisy circles (see Fig. 1a and 1d). In order to remove the ring artifact, the image transformed to polar coordinate space is subjected to the 2D Fourier transformation. Then, the strip shaped noises are expressed as high frequency component. It is noted that the image in Fig. 3a is separated into the half and stitched together to have a better transformation. The, the high frequency is removed and the inverse transformation results in the image whose strip noises are removed. It is highlighted that the anisotropic diffusion and median filter may create enhanced images as well with less clarity(Perona and Malik, 1990). Then, the transform into the Cartesian coordinate results in the image shown in Fig 4a. The phase separation (Fig 4b) then becomes clearer than that before enhancement.

0 100 200 300 400 5000

1000

2000

3000

4000

5000

0 100 200 300 400 5000

1000

2000

3000

4000

5000

θ

r

r

Mea

n pi

xel v

alue

4th polynomial Function, Fm(r)

Ic

center periphery r

Mea

n pi

xel v

alue

periphery center

(a) (b)

(c)

. Fig 3. (a) Stitched image. (b) Frequency spectrum. (c) Noise-removed image in Polar coordinate space.

Fig 4. (a) Image where beam hardening and cupping effects are removed. (b) Segmented binary image where black dots indicate air void.

2.3 Separation of Multi-phase

The segmentation procedure is straightforward by defining a threshold value. Most geomaterials of interest are comprised of solid and void so that the x-ray attenuation values have a wide distribution due to its porosity. The most common and readily applicable method is to minimize intra-class variance via Otsu’s method(1979). This algorithm is well implemented in the ImageJ and MatLab. However, this type of method may not be feasible when it comes to more than 3 phase materials (i.e., unsaturated

High frequency

θ

r

soil) as sIt is notdistributmanifest (a)

F

The multhis casphase wair is nowater phvalue ofWhat wapproximphase inend is pphase fimpleme (a) Soli

shown in Ftable that ion in poret for each

ig 5. (a) 2D

ltiple segme. Howeve

whose peakot sharp ehase. Thef threshold

we implemmate waten Fig 6b shpresented ifraction (eentation.

id

F

Fig 5a. Thewater pha

e. Fig 5b shphase and

D sliced im

mentation aer, it fails tk exists at

enough so refore, all

d although ented in tr phase bhould be pn Fig 7 whe.g., poro

Fig 6.(a) S

e black colase form hhows the d

d a large vo

(b)

mage of uns

algorithmsto approprthe middlethat pixel the particit become

this study ecause thart of both

here three osity, degr

(b) Wate

olid phase

or denotesheterogenedistributionolume of m

saturated s

should beriately sege (water ph

values arle boundaes a first a

is iterative pixels a

h solid andphases arree of sa

er

e. (b) Wate

s air, whiteeous patch

n of pixel vamixel exist b

soil. (b) His

e able to dement phas

hase). Theround partries are d

and good sve dilationat particle air. Resul

re nicely seaturation)

r phase. (c

e for solid, ah rather thalues. Threbetween p

stogram of

efine two tses in Fig boundary

ticles are setected bystart of ph

n and erosboundary tant phase

egmented. may be

(c) Air

c) Air phas

and grey fhan homoee peak vahases.

f pixel valu

threshold v6, in part

between ssame as ty applying hase segmsion of phcaptured a

e distributio It is notedvarying w

e

for water. geneous

alues are

ue.

values in icular for solid and those for a single entation. hases to as water on at the d that the with this

3. Pore

Delaunanearest Raoush as showevolutioninformatbe readwhich pr

Character

ay Tessellaparticle ceet al. (200

wn in Fig 8.n of pore tion in poroily obtainarovide pore

Fig 7

rization

ation dividenters. It te03) sugges. The identcharacteris

ous media able by ape path and

. Segment

Fig 8. De

es the spaends to ovested definintification ofstics with is the netw

pplying bot the junctio

ted unsatu

efinition of

ace into trer-divide png mergingf unit pore strain andwork of porth thinningon.

rated soil i

unit pore

riangle (tepore spaceg criteria frhelps eva clusteringre structure

g and mod

A

W

S

mage

trahedron es with sevrom inscribluating por

g of particle. The skedified Dela

Air

Water

Solid

in 3D) coveral unit pbing spherere size disles. Anotheeletal strucaunay Tess

onnecting pores. Al-e of pore tribution, er useful ture may sellation,

Fig. 9. C

As the algorithmpore netdistribut 4. CONC The combecomeunless aseveral artifact),Most imengineeobjective REFERE Al-Raou

techn1700.

Characteriz

Fig. 10.

fitting of im permits twork as illion of pore

CLUSIONS

mputed toms a powerappropriateimage pro

, phase seage enhanring, mathe of image

ENCES

ush, R., Thniques for u.

zation of po

Network o

nscribing stracking plustrated ine chamber

S

mography rful non-dee image pocessing e

eparation, incement tehematics ae processin

hompson, Kunconsolid

ore networ

of pore cha

sphere appore channn Fig. 10. Band chann

naturally pstructive te

processing examples dentificatio

echniques nd granula

ng and to s

K., and Wated porou

rk by thinni(MDT)

amber and

pproximatenel (or poreBased on tnel.

provides inesting tooltechniquefor noise

on of unit pused in thar physicsselect suita

illson. C. Sus media.”

ing and mo

interconne

es the poree throat), ithis effort,

nternal str. Yet, it re

es are invoreduction

pore and cis study ar. It is how

able metho

S., “Compa”, Soil Sci.

odified Del

ected pore

e chambet enables we quantit

uctures ofsides in a olved. This (beam hconstructiore availableever signifdologies.

arison of nSoc. Am.

aunay Tes

e channel

r and pathcreating th

tatively ana

f geomatequalitative

s paper inhardening on of pore e in medicficant to c

network geJ, 67 (200

ssellation

h finding he entire alyze the

rials and e domain troduces and ring network.

cal image larify the

eneration 03) 1687-

Barrett, J. F., and N. Keat., "Artifacts in CT: Recognition and Avoidance1." Radiographics , 24 (2004) 1679-1691.

Kikuchi, Y., Hidaka, T., Sato, T., and Hazarika, H., “Deformation characteristics of tire chips-sand mixture in triaxial compression test by using X-ray CT scanning.”, Advances in computed tomography for geomaterials, GEOX 2010, New Orleans, LA, USA, (2010) 67-75.

Otsu, N., “A Threshold Selection Method from Gray-Level Histograms, Systems, Man and Cybernetics”, IEEE Transactions, 9 (1979) 62-66.

Perona, P., and Malik, J., “Scale-space and edge detection using anisotropic diffusion.”, IEEE Transactions on Pattern Analysis and Machine Intelligence, 12 (1990) 629-639