Bollettino di Geofisica Teorica ed Applicata Vol. 61, n. 2, pp. 145-158; June 2020
DOI 10.4430/bgta0303
145
VSP wavefield separation using structure tensor dip masking filter
H. HasHemi and m. GHasem FakHari
Institute of Geophysics, University of Tehran, Iran
(Received: 19 June 2019; accepted: 18 October 2019)
ABSTRACT InaVerticalSeismicProfiling(VSP)acquisition,downgoingwavefieldsandupgoingwavefields, interfere. Both wavefields are practical in various seismic applications,However,itisneededtoseparatethesetworeflectionandtransmissionfields.DifferenttechniqueshavebeenproposedforVSPwavefieldseparation.Medianfilteringaswellas2DFourier transformsarecommonlyused for separating the seismicwavefields.The former suffers from the averaging effect, generating artifacts, and amplitudemodification due to spectral energy leakage after the inverse transform. 2DFourierhastheproblemofleakageandedgeeffects.WeproposeanewapproachbasedonthestructuretensorandlocaldipestimationfollowingamaskingfilterforVSPwavefieldseparation.Thedipmaskingfilter is calculatedusing the local dip of eachpoint ofdatathatcanseparateupgoinganddowngoingwavefieldsfromtheoriginaldata.Theadvantageofthisapproachisbothinpreservingseismicamplitudesandinprovisionofasectionfreeoffakeevents;thisliterallyimpliesnoenergyleakage.Thesyntheticandrealdataexamplesaredemonstratedtoshowtheperformanceoftheproposedmethod.
Key words: VSP,waveseparation,localdip,structuretensor,dipmaskingfilter.
© 2020 – OGS
1. Introduction
VerticalSeismicProfiling(VSP)isaseismictechniquethatextractshigh-resolutioninformationfromthesubsurface.TheseismicVSPsignal isgeneratedat thesurfacerecordedbyreceiverslocatedatdifferentdepthsinaverticalwell.VSPrecordeddatahavebeenusedincalculatingthevelocityandtheattenuationinaninterval(Pevzneret al.,2011;BaharvandAhmadiandMorozov,2013;Wang,2014).
InaVSPrecording,downgoingwavefieldswithapositive-slownessandupgoingwavefieldswithanegative-slowness,interferewitheachother(SeemanandHorowicz,1983;Hardage,2000).ThesimultaneousrecordingofupgoinganddowngoingwavefieldsareregardedasanaddedvalueoftheVSPmethod.DowngoingwavesareusedtocomputeseismicP-andS-wavesvelocities,tocalculatetheseismicanelasticqualityfactorQ,tocreatetime-dependentfilters,whichareusedtorecoverhigh-frequencyseismicwaves(SudhakarandBlias,2002).UpgoingwavefieldsaretoolsforobtainingVSP-CDPimages(Blias,2005).
AsafundamentalsteptoprocessaVSPdataset,theseparationofupgoinganddowngoingwavefieldsisintroduced.Therearemanymethodstoseparateupgoinganddowngoingsignalsfromeachother.Mostlythesemethodsarebasedontwotypes:1)methodsusingaveragingfilters
146
Boll. Geof. Teor. Appl., 61, 145-158 Hashemi and Ghasem Fakhari
for the separationof theupgoing anddowngoingwavesbymedianfilter; 2)wave separationtechniqueswhichtransformprimarydataintoanewdomainandseparateupgoinganddowngoingwavesinthisspaceandthereafterreturningdatatoitsprimaryspace.Themostimportantonesaref-kandτ-pfiltering.Curveletsarealsoakindofdirectional-scaletoolsusedforVSPwavefieldseparation(Heraviet al.,2012).
The median filter method (as a conventional wave separation technique) is based onflattening data for separating downgoing, then applying a classicmedian filter, shifting backandfinallysubtractingtheseparateddowngoingwavefieldfromtheoriginaldatatoobtaintheupgoingwavefields.Inthemedianfiltermethod,eachsampleisreplacedbythemedianvalueofitsneighbourhoodsamples.Therefore,duetotheaveragingeffect,thismethodsuffersfromsmoothingtheinputsignal,manipulatingprimarysignalamplitudes,anddamagingsomeoftheusefuldetails.Tocopewiththeseproblems,someimprovedversionsofmedianfiltersornon-linearmedianfiltershavebeenproposedsuchasvectormedianfiltering(KasparisandEichmann,1987;Astolaet al.,1990)whichhasbeenappliedinseismicdataprocessing(Liu,2013).Moreover,Chenet al. (2016) improved thesignalpreservingabilityofamedianfilterusingastructure-orientedspace-varyingmedianfilter.
Waveseparationtechniquessuchasf-kortheτ-p,transformdatainthenewdomainandalwaysgeneratingsomeartifactseventsinthisspaceisunavoidableandhenceamplitudemodificationduetospectralenergyleakageeffecthappens.Inexperimentalphysics,dataspectralleakageisawell-knownprobleminthedatadomaintransforms.Xuet al.(2005)proposedanantileakageFouriertransformapproachforseismicdataregularisationcase.Chenet al.(2014)proposedaniterative framework for deblending of simultaneous-source seismic data usingSeislet domainshapingregularisationandcomparedthismethodwithtwoFourierbasedtransform;f-kdomainthresholding, and f-x predictive filtering.According to their results, Fourier based transformmethodssufferfromenergyleakageandgenerateartifacts.
DowngoingwavesarecharacterisedbypositiveapparentvelocityandanegativeslopeinVSPimagedata,upgoingwavesarecharacterisedbynegativeapparentvelocityandpositiveslope.Inthispaper,itisdecidedtousethedipvarianceand,then,theoutputisusedtoseparateupgoinganddowngoingwavefieldsfromeachother.So,datawillnotbetransferredtothenewspaceandasaresult,datadomaintransferdoesnotyieldtheleakageofenergy.Inthispaper,wedecidetomakeamaskingfilterbasedonthelocaldipofeachpointofdatawhichcanseparateupgoinganddowngoingwavefieldsfromoriginaldata.Wecalledthismethod“dipmaskingfilter”andtheadvantageisthatneitheraveragingofamplitudesnorartificialeventbytransferdatahappen;thisimpliesnoenergyleakage.
An initial published work estimating dip directly from 2D seismic data is by Picou andUtzmann(1962).FinnandBackus(1986)extendeddipestimation to3Ddataasapiecewisecontinuousfunctionofspatialpositionandseismictraveltime.Marfurtet al.(1998)generalisealatersemblance-basedscanbyFinnandBackus(1986).AsdiscussedbyMarfurt(2006),thelocal reflection orientation can be described by reflection normal vector. Fomel (2002) usedplane-wavedestructionfiltertocomputereflectionslopes.vanVlietandVerbeek(1995)presentanestimatebasedonthegradientstructuretensor.ChenandMa(2014)proposedadip-separatedfiltering using an adaptive empiricalmode decomposition based on dip filter to separate theseismicdataintoanumberofdipbands.Thismethodsaysthatthedipestimationisbetterwhenitisappliedtodip-separatedprofiles.
VSP wavefield separation with structure tensor Boll. Geof. Teor. Appl., 61, 145-158
147
Ourproposedlocaldipmaskingfilter,basedonthestructuretensormethod,isusedforthepurposeofcalculatingdipvariance.Thecoherentstructureofseismicimagesyieldstosupposegoodcandidatesforstructuretensorapplication(vanVlietandVerbeek,1995;Weickert,1999;FehmersandHöcker,2003),whichareacommontooltoestimatethelocalorientationofimagefeatures.Tensorstructureinseismicsisusedinestimatingtheorientationsoftheseismicstructuralcharacteristicssuchasreflectionslope,stratigraphicfeaturesorientations,horizoninterpretation,subsurfacemodelling,etc. (e.g.LiandOldenburg,2000;Bakker,2002;FehmersandHöcker,2003;Marfurt,2006;Hale,2009;WuandHale,2015a;Wu,2017a,2017b).Bakker(2002)andWuandHale(2015b)usestructuretensors(e.g.vanVlietandVerbeek,1995;Weickert,1997)toestimatereflectionorientations.WuandJanson(2017)proposeadirectionalstructuretensortoestimatetheorientationsofreflectionswithsteepandrapidlyvaryingslopes.
FarakliotiandPetrou(2005)benefitfromtheuseofstructuretensorstoanalysetheseismicdata.Inthispaper,thestructuretensoralgorithmisusedtoseparateVSPwavefieldsimagesafterfindingtheirlocaldips.
2. Methodology
2.1. Estimating local dips using structure tensorsThestructuretensoristheouterproductoftheimagegradientwithitself,whichisalsocalled
gradient-squaretensors.Structuretensorwasdevelopedprimarilyforedgedetection(FörstnerandGülch,1987),butithasbeenusedforawiderangeofproblemsinimageprocessing.
ConsideringanimageI(x.y)thestructuretensorSisdefinedas:
(1)
whereK is a smoothing kernel (e.g aGaussian kernel),* stands for convolution operator, Ix and Iy are theverticalandhorizontalcomponentsof thegradientof the image I, respectively.In2D,thestructuretensorSisfinallyofn×m×2×2wheren×misthesizeoftheinitial imageI.Effectively,tensorstructureobtainsa2×2matrixateachpixeloftheoriginalimage.BytheeigendecompositionofthematrixS,wecanobtaintheorientationinformationineachpixel.Inaccordancewiththeprinciplesofmatrixeigenvectordecomposition,theeigendecompositionofa2Dstructuretensorisasfollows:
(2)
S = λu uuT + λv vvT. (3)
where u and v are normalised eigenvectors corresponding to the eigenvalues λu, λv,respectively.
These eigenvectors provide an approximation of orientations of features in the 2D image(FehmersandHöcker,2003).Thisinformationcanbevisualisedasanellipsewithtwodiameters
148
Boll. Geof. Teor. Appl., 61, 145-158 Hashemi and Ghasem Fakhari
whichareequaltotheeigenvaluesanddirectedalongtheircorrespondingeigenvectorsasshowninFig.1.
AsshowninFig.1, theorthogonaleigenvectorsuandvdescribetheorientationoffeaturein each point. Individually, for each sample, the eigenvector u, corresponding to the largesteigenvalue(λu),isparalleltothedirectionsinwhichtheimagefeaturesvarymostsignificantly.AsshowninFig.1,thecomponentsoftheunitvectorûarerelatedtolocaldipsθineachsampleby:
u1 = |u| cos θ, (4)
u2 = |u| sin θ, (5)
Finally,thedominantorientation(localdip)ineachsampleiscomputedfromtheeigenvectoruassociatedwithλuas:
dip(0)=tan–1(u2/u1). (6)
2.2. Wave separation process using dip masking filterUsing dip masking filter for VSP wavefield separation based on the structure tensor, the
followingstepsareproposed:1. applicationofasmoothingfilter(e.g.withGaussiankernel)ondatatoreducerandomnoise
andnon-structuralorientations;2. calculatethepartialderivativesoftheimageandbuildstructuretensor;3. eigenvaluedecompositionofthestructuretensor;4. estimatelocaldipovereachsample;5. useamovingsimilarityfilterondipimagetoobtainthedominantslopeofeachsample;6. applyamediandipfilterondatatoeliminatenoisepointswithnon-structuraldirections;
Fig.1 -Thestructure tensoratpointOis visualised as an ellipse and its uniteigenvectorsu,vandrootedeigenvaluesλu,λvarealsodepicted.
VSP wavefield separation with structure tensor Boll. Geof. Teor. Appl., 61, 145-158
149
7. createtwodipmaskingfiltersbyseparatepositiveandnegativeobtaineddips;8. apply created dip masking filters on VSP data to separate upgoing and downgoing
wavefields;9. usingthef-kinterpolationtorecoveringintersectingpointsafterapplymaskingfilter.
Inanoise-freedata,theestimatedorientationbystructuretensorwillbeanidealapproximationfortheslopeoftheeventsreportedinsamplebysamplemanner.But,ingeneral,thepresenceofnoiseinthedatawillaffecttheestimateddirections.Inpractice,mixingthecontentofcoherentsignal and random noise in seismic data is unavoidable. Therefore, we require to computemore stabledominantorientationsby implementing a smoothingfilter to each elementof thegradient-basedstructuretensors.Thesmoothinghelpstoremovethenoiseandmuchmorestableorientationsevenifapparently, theresolutionof theinput imageisnotgood.Inthealgorithmpresented,thefirststep,relatedtosmoothingdatabeforecalculatingthedip,isdonewithseriousconsiderations.WeproposedarecursiveGaussianfilterforsmoothingtheparametersofwindowwidths both in time and spatial directions, thesemust be optimally chosen for each input. Itsmoothesafewpixelswhosevaluesdiffersignificantlyfromtheirneighborhoodwithoutaffectingtheotherpixels.Notealso thatby increasing thesizeof the smoothingwindow, the structuretensorisrobustinthepresenceofnoisydatawithlesseffectinreducingthespatialresolution[formoredetailsaboutGaussianfilterandchoosingoptimumsmoothingparameterrefertoDeriche(1992),vanVlietet al.(1998)andHale(2006,2009)].
Steps2,3,and4arerelevantincalculatinglocaldipovereachsampleusingstructuretensorsmentionedinsection2.1.Beforecreatingmaskingfiltersfromtheestimatedlocaldips,inadditiontothesmoothinginstep1,thealgorithmperformsthetwoadditionalsteps5and6todealwithnoiseandestimatedrandomorientations.
Step5issomewhatsimilartothenonlocalalgorithmfortheimagedenoising,exceptthatitdoesnotrefertotheimageandisonlyappliedtoestimateddirectionsfromstructuretensors.Aftercomputationoflocaldirectionsineachsample,amatrixwith2×1elementisconstructedateachpixeloftheoriginalimage.Next,thismatrixisdecomposedintoanumberofvertical-horizontaloverlappingsmallwindows.Forthecentralpixelofeachwindow,thecosinesimilaritybetweenthispixelandneighbourhoodsiscalculatedandonthebasisofthereportedvalue,itisdecidedtokeep/removethedirectionforthatpixel.Athresholdsimilarityvaluebytheusermustbesetinthealgorithm.TheschematicperformanceofthisstepisshowninFig.2.
Fig.2-Schematicperformance ofmovingsimilarityfilterpresentedinstep5onthedipimagetoobtainthedominantslopeofeachsample.
150
Boll. Geof. Teor. Appl., 61, 145-158 Hashemi and Ghasem Fakhari
Instep6,amedianfilterisappliedondatausingtheestimateddominantdipinformation,toeliminatenoisepointswhichhavenon-structuraldirections. Instep7, twodipmaskingfiltersarecreatedbyseparatepositiveandnegativecalculateddipsandthemaskingfiltersareappliedaccordinglyonVSPdata toseparateupgoinganddowngoingwavefields instep8.Finally, inthelaststep,thef-kinterpolationisusedtorecoverintersectingpointsafterapplicationofthemaskingfilter.
Ingeneral,ourproposedmethodisnotanoiseremovalmethodand,ifthereissignificantnoiseindata,itisbettertousedenoisingprocessesbefore.
3. Examples
3.1. Application on a synthetic modelTodisplaytheefficiencyoftheproposedmethod,itisfirstlytestedonasyntheticVSPdata
set.Fig.3ashowsasimplesyntheticVSPdatasetcontainingasingledowngoingandasingleupgoingwavefield.Fig.3bshowseigenvectorscomputedfromstructuretensorsineachsample
Fig.3-a)AsimplesyntheticVSPdatasetwithasingleupgoingandasingledowngoingwavefield;b)eigenvectorscomputedfromstructuretensorsineachsampleexhibitinglocalstructuralorientation;c)andd)VSPdowngoingandupgoingwavesobtainedfromapplying dipmaskingfilteronsyntheticdata.
VSP wavefield separation with structure tensor Boll. Geof. Teor. Appl., 61, 145-158
151
where structuralorientations in thesepoints are shown (for abettervisualisation,1outof20eigenvectorsinallimagesisshown).
After the creation of two dipmasking filters, namelywith positive and negative obtaineddips,andperformingthedipmaskingfilteronsyntheticVSPdata,theseparateddowngoingandupgoingwavefieldsareobtained(Figs.3cand3d,respectively).AsitisshowninFigs.3cand3d,theproposedmethodseparatesdowngoingandupgoingwavefieldsfromeachotherwithoutartifactsorsmoothing.
AsshowninFig.3b,inthisimageall theorientationsarecorrectlyestimatedbystructuraltensorandtheycontainthreedominantdirections:1)adowngoingeventwithnegativedip,2)upgoingeventwithpositivedip,and3)background.
Nowweadd theGaussianwhitenoise to theprimarysyntheticVSPdata setandcomputeeigenvectorscalculatedfromstructuretensorswithandwithouttheuseofasmoothingfilter.TheresultsforthetwonoisevaluesareshowninFigs.4and5.
Fig.4-a)NoisysyntheticVSPdatasetcorruptedbyamiddlelevelofGaussianwhitenoise;b)eigenvectorscomputedfromstructuretensorswithouttheuseofasmoothingfilter;c)eigenvectorscomputedfromstructuretensorswiththeuseofaGaussiansmoothingfilter.
AsshowninFigs.4and5,thenoiseaffectstheestimateddirectionsandcausesthedeviationoftheslopeapproximationoftheevents.But,byusingthesmoothingfilteronbothimages,theestimatedorientationsfordominantdirections(upgoinganddowngoing)arecorrectlycalculated.
152
Boll. Geof. Teor. Appl., 61, 145-158 Hashemi and Ghasem Fakhari
3.2. The issue of intersecting wavesStructuraltensorsareagoodestimatorforfindingthedirectionofthewavepaths,butonly
forcases inwhich theydonot intersect.Hence, thestandardstructural tensorbreaksdown inthe presence of intersecting wavefields. In order to solve the difficulty in estimating correctconflictingdips,Chen(2016)proposedadip-separatedfilteringapproach.However,inthisstudyourobjectiveistoseparateupgoinganddowngoingwavefieldsinthesimplestwaywiththelowestcost;itimpliesthatthepreciselocalslopeestimationattheintersectionpointisnottargeted.
For the case of two intersecting local orientations with different directions, the proposedapproachcansmartlyrepresentonlyoneofthedominantpathsforthecorrespondingdirectionusingf-kinterpolation.
InFig.6a,asimpleexampleoftwowaveswithdifferentlocalorientationsisshown.Figs.6band6cshowthecomputedlocaldipvectorsfromstructuretensorsineachsample.Asshowninthisfigure,attheintersectionpointonlythemostdominantdirectionisrestored.Aftercalculatingthelocaldipateachpoint,theupgoinganddowngoingmaskingfiltersarecreatedbycollectingpositiveandnegativedips,respectively.Finally,thedipmaskingfiltersareappliedontheinputdata(Fig.6a)andtheseparatedupgoinganddowngoingwavefieldsaregenerated(Figs.6dand6e,respectively).AsshowninFig.6c,onlytheupwarddirectionisobtainedattheintersectionpoint,soinfindingthemaskingfilter,thisareaisassignedonlytoupgoingwavefield.Therefore,
Fig.5-a)NoisysyntheticVSPdatasetcorruptedbyahighlevelofGaussianwhitenoise;b)eigenvectorscomputedfromstructuretensorswithouttheuseofasmoothingfilter;c)eigenvectorscomputedfromstructuretensorswiththeuseofaGaussiansmoothingfilter.
VSP wavefield separation with structure tensor Boll. Geof. Teor. Appl., 61, 145-158
153
the nearby samples are incorrectly removed in downgoing wavefields and a discontinuity isobservedindowngoingwavefieldatthisintersectingpoint(Fig.6e).
In this paper, we use the f-k interpolation technique to recover intersecting points afterapplyingamaskingfilterondata.Fig.6fshowsthedowngoingwaveafterrecoveringbythef-kinterpolationattheintersectionpoint.
3.3. Real data exampleWeconsidernowarealverticalseismicprofilefroma3Delasticsimulation.Thisdataset
hasbeenacquiredinSEAMPhaseIRPSEAbasedonthegeologyoftheGulfofMexico.Thedescribedmethodgivesusatooltoseparatedowngoingandupgoingwavefields.Fig.7showsthisVSPdatasetconsistingofdowngoingandupgoingwavefields.
Fig.6-a)Aasimpleexampleoftwowaveswithdifferentlocalorientations;b)eigenvectorscomputedfromstructuretensors in each sample; c) eigenvectors in each point; d) upgoing events obtained from the application of the dipmaskingfilteronprimarydatashowninpanela;e)downgoingeventsobtainedfromtheapplicationof thedipmaskingfilteronprimarydatashowninpanela,attheintersectionpoint,adiscontinuityisobservedindowngoingwavefield;f)downgoingwavefieldafterrecoveringbythef-kinterpolationattheintersectionpoint.
Fig. 7 - A real vertical seismic profile data setconsistingofdowngoingandupgoingwavefields.
154
Boll. Geof. Teor. Appl., 61, 145-158 Hashemi and Ghasem Fakhari
Fig. 8 shows eigenvectors computed from structure tensors in each sample where there is astructuralorientationinthesepoints.Thesedominantdirectionshelpustoseparatedowngoingandupgoingeventsfromeachother.
Fig.8-Eigenvectorscomputedfromstructuretensorsineachsample.Thedirectionoftheseeigenvectorscanrepresenta localstructuralorientationineachsample(ofcourse,forthepresentation, theeigenvectorofall thepointsisnotdrawnbutonlyoneouttenpoints).
AfterapplyinggenerateddipmaskingfiltersontherealVSPdatashowninFig.7,theseparateddowngoingandupgoingwavefieldsaregenerated(Figs.9aand9b,respectively).
Fig.9showstheprogressionobtainedbythedipmaskingfilterintheupgoinganddowngoing
Fig. 9 - Downgoing events obtained from applying the dipmasking filter on primaryVSP data (a) and upgoing events(b).
VSP wavefield separation with structure tensor Boll. Geof. Teor. Appl., 61, 145-158
155
separation.As it is clear in this figure, in points of discontinuity intersection in the upgoinganddowngoingeventscontinuityofeventsislost.Thesolutionpresentedhereistheuseoff-kinterpolationinintersectingpoints.
Figs.10aand10bare the separateddowngoingandupgoingwavefieldsafter interpolationattheintersectingpoints.Thesefigureshighlightthatthecontinuityofseismiceventshasbeensufficientlyrecovered.Theresultofanotherstandardwavefieldseparationmethod(flatteningandsubsequentmedianfiltering)isshowninFigs.11aand11b.
Comparingtheseparationresultsobtainedbytheapplicationofthemedianfiltermethod(Fig.11)with theproposedmethod (Fig.10), it is clear that there isnot significant interferenceof
Fig. 10 - Downgoing separated wavefields recovered in intersection points by f-k interpolation (a) and upgoingwavefieldsafterrecovered(b).
Fig.11-Downgoingseparated(a)andupgoingwavefieldsobtainedbymedianfiltermethod(b).
156
Boll. Geof. Teor. Appl., 61, 145-158 Hashemi and Ghasem Fakhari
downgoingeventsintheseparatedupgoingwaves(Fig.10b),butclearly,downgoingwavesarevisibleinseparatedupgoingwavefieldobtainedbythemedianfiltermethod(Fig.11b).
Comparing the separationdowngoing results, it appears thehighly coherent visible outputof standardmedian filtering algorithm. However, a further look to the computed histogramsof amplitudes (Fig.12) confirms that themedianfilteringhasconsiderablychanged thenormof amplitudevalues anddistributionconsiderably.The similarityofFigs. 12a and12bhighlyemphasisestheamplitudepreservingmanneroftheproposedmethod.
Furthermore,firstbreakpickingisafundamentalstepinVSPwavefieldseparationbymedianfiltermethod,anyerrorsofthesearrivaltimesmayhavesignificanteffectsontheresults.Anotheradvantageoftheproposedmethodisthatitdoesnotrequirethesetimes.
Fig.12-Histogramofamplitudeof:a)originalVSPdata;b)downgoingwavefieldsobtainedbydipmaskingfilter,andc)downgoingwavefieldsobtainedbymedianfiltermethod.
VSP wavefield separation with structure tensor Boll. Geof. Teor. Appl., 61, 145-158
157
4. Conclusion
Inthispaper,wehaveproposedanewapproach,structuretensorfollowedbydipmaskingfilter, to separate VSP downgoing and upgoing waves. The main idea of this method is theapplicationofslopeofeventstoseparateupwardanddownwardwaves:wefirstgenerateamaskwiththesamesizeofdatausingtheslopesderivedfromstructuretensor,then,byapplyingthemaskfilterondata,weseparatetheupwardanddownwardwaves.Thisleadstonoenergyleakage(amplitudepreservingmanner)andthetotalenergyafterseparationisequaltotheinitialinputensembleenergy,despite that,intraditionalmethods,waveseparationoftenoccurswhendataistransferredtoanewspace(f-kdomainorτ-pdomain).
Finally, itwasshownthat theapproachpresentedinthispaperperformswell inseparatingVSPdowngoingandupgoingwavesandofferstheadvantageoverconventionaltechniquesthatnoaveragingofamplitudeofdataoccurs,sonoartificialsmoothingoftheeventisdoneandhencetheleakageofenergyisnearlyzero.
REFERENCES
AstolaJ.,HaavistoP.andNeuvoY.;1990:Vector median filters.In:Proc.IEEE,SpecialissueonMultidimensionalSignalProcessing,vol.78,no.4,pp.678-689.
BaharvandAhmadiA.andMorozovI.;2013:Anisotropic frequency-dependent spreading of seismic waves from first-arrival vertical seismic profile data analysis.Geophys.,78,C41-C52,doi:10.1190/geo2012-0401.1.
BakkerP.;2002:Image structure analysis for seismic interpretation.PhDthesisinAppliedPhysics,DelftUniversityofTechnology,theNetherlands,120pp.
BliasE.;2005:VSP wavefield separation, wave-by-wave approach.Geophys.,72,T47-T55,doi:10.1190/1.2744124.ChenY.;2016:Dip-separated structural filtering using seislet transform and adaptive empirical mode decomposition
based dip filter.Geophys.J.Int.,206,457-469,doi:10.1093/gji/ggw165.ChenY.andMaJ.;2014:Random noise attenuation by fx empirical-mode decomposition predictive filtering.Geophys.,
79,V81-V91,doi:10.1190/GEO2013-0080.1.ChenY.,FomelS.andHuJ.;2014: Iterative deblending of simultaneous-source seismic data using seislet-domain
shaping regularization.Geophys.,79,V179-V189,doi:10.1190/GEO2013-0449.1.ChenY.,ZuS.,WangY.andChenX.;2016:Deblending of simultaneous source data using a structure-oriented space-
varying median filter.Geophys.J.Int.,216,1214-1232,doi:10.1093/gji/ggy487.DericheR.;1992:Recursively implementing the Gaussian and its derivatives.In:Proc.,SecondInt.Conf.onImage
Processing,Singapore,pp.263-267.FarakliotiM.andPetrouM.;2005:The use of structure tensors in the analysis of seismic data. In:IskeA.andRandenT.
(eds),Mathematicalmethodsandmodellinginhydrocarbonexplorationandproduction,Mathematicsinindustry,Springer,Berlin,Heidelberg,Germany,vol7.,pp.47-88,doi:10.1007/3-540-26493-0_3.
FehmersG.C.andHöckerC.F.;2003:Fast structural interpretation with structure-oriented filtering.Geophys.,68,1286-1293,doi:10.1190/1.1598121.
FinnC.J.andBackusM.M.;1986:Estimation of three-dimensional dip and curvature from reflection seismic data.SEGTechnicalProgramExpandedAbstracts,pp.355-358,doi:10.1190/1.1893089.
FomelS.;2002:Applications of plane-wave destruction filters.Geophys.,67,1946-1960,doi:10.1190/1.1527095.FörstnerW.andGülchE.;1987:A fast operator for detection and precise location of distinct points, corners and
centres of circular features.In:Proc.,ISPRSIntercommissionConferenceonFastProcessingofPhotogrammetricData,Interlaken,Switzerland,pp.281-305.
HaleD.;2006:Recursive Gaussian filters.CenterforWavePhenomena,ColoradoSchoolofMines,Golden,CO,USA,CWPReport546,10pp.
HaleD.;2009:Structure-oriented smoothing and semblance.CenterforWavePhenomena,ColoradoSchoolofMines,Golden,CO,USA,CWPReport635,10pp.
HardageB.A.;2000:Vertical seismic profiling, 3rd updated and revised edition. In:HelbigK.andTreitelS.(eds),HandbookofGeophysicalExploration,SeismicExploration,Pergamon,NewYork,NY,USA,vol.14,572pp.
158
Boll. Geof. Teor. Appl., 61, 145-158 Hashemi and Ghasem Fakhari
HeraviM.F.,MohamadiH.andHashemiH.;2012:A curvelet based wavefield separation in vertical seismic profiling.In: SEGGlobalMeetingAbstracts, InternationalGeophysicalConference andOil&GasExhibition, Istanbul,Turkey,pp.1-4,doi:10.1190/IST092012-001.35.
KasparisT.andEichmannG.;1987:Vector median filters,SignalProcess.,13,287-299.LiY.andOldenburgD.W.;2000:Incorporating geological dip information into geophysical inversions.Geophys.,65,
148-157.LiuY.;2013:Noise reduction by vector median filtering.Geophys.,78,V79-V87,doi:10.1190/geo2012-0632.1.MarfurtK.J.;2006:Robust estimates of 3D reflector dip and azimuth.Geophys.,71,P29-P40,doi:10.1190/1.2213049.MarfurtK.J.,KirlinR.L., Farmer S.L. andBahorichM.S.; 1998:3-D seismic attributes using a semblance-based
coherency algorithm.Geophys.,63,1150-1165,doi:10.1190/1.1444415.PevznerR.,GurevichB.andUrosevicM.;2011:Estimation of azimuthal anisotropy from VSP data using multicomponent
S-wave velocity analysis.Geophys.,76,D1-D9,doi:10.1190/geo2010-0290.1.PicouC.andUtzmannR.;1962:La “coupe sismique vectorielle”.Geophys.Prospect.,10,497-516.SeemanB.andHorowiczL.;1983:Vertical seismic profiling: separation of upgoing and downgoing acoustic waves in
a stratified medium.Geophys.,48,555-568,doi:10.1190/1.1442170.SudhakarV.andBliasE.;2002:Transfer function method of seismic signal processing and exploration.U.S.Patent,
US6,340,508B1,11pp.vanVlietL.J.andVerbeekP.W.;1995:Estimators for orientation and anisotropy in digitized images.In:Proc.,1st
ConferenceAdvancedSchoolforComputingandImaging(ASCI’95),Heijen,TheNetherlands,pp.442-450.vanVlietL.J.,YoungI.T.andVerbeekP.W.;1998:Recursive Gaussian derivative filters.In:Proc.,14thInternational
ConferenceonPatternRecognition(ICPR),Brisbane,Australia,vol.1,pp.509-514,doi:10.1109/ICPR.1998.711192.WangY.;2014:Stable Q analysis on vertical seismic profiling data.Geophys.,79,D217-D225,doi:10.1190/geo2013-
0273.1.WeickertJ.;1997:A review of nonlinear diffusion filtering.In:RomenyB.t.H.,FlorackL.,KoenderinkJ.andViergever
M.(eds),Scale-SpaceTheoryinComputerVision,LectureNotesinComputerScience,vol.1252,Springer,Berlin,Heidelberg,pp.3-28.
Weickert J.; 1999: Coherence-enhancing diffusion filtering. Int. J. Comput. Vision, 31, 111-127, doi:10.1023/A:1008009714131.
WuX.;2017a:Building 3D subsurface models conforming to seismic structural and stratigraphic features.Geophys.,82,IM21-IM30,doi:10.1190/geo2016-0255.1.
WuX.; 2017b:Directional structure-tensor-based coherence to detect seismic faults and channels. Geophys.,82,A13-A17,doi:10.1190/geo2016-0473.1.
WuX.andHaleD.;2015a:Horizon volumes with interpreted constraints.Geophys.,80,IM21-IM33,doi:10.1190/geo2014-0212.1.
WuX.andHaleD.;2015b:3D seismic image processing for unconformities.Geophys.,80,IM35-IM44,doi:10.1190/geo2014-0323.1.
WuX.andJansonX.;2017:Directional structure tensors in estimating seismic structural and stratigraphic orientations.Geophys.J.Int.,210,534-548,doi:10.1093/gji/ggx194.
XuS.,ZhangY.,PhamD. andLambaréG.; 2005:Antileakage Fourier transform for seismic data regularization. Geophysics,70,4,V87-V95.
Corresponding author: Hosein Hashemi Institute of Geophysics, University of Tehran North Kargar street, Tehran, Iran Phone: + 98 21 88001115; e-mail: [email protected]