版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领
文档简介
图像割图像析计机视第重要最的骤之,也认为机定位和区域增长,这些为更快的图像/分析和识别系统的发展提供了很好的机(FSFS(缘检测算法也包括在这个程序步骤中。分割算法利用了边缘信息和平滑后的图像,用来找到图像中的区域。在这项工作中FS分割算法被选择是依赖于它在一些列应用中((T用StrtchS5000用它的性能。有一些分割方法,它们不在RGB空间里分割彩像,因为它们和HISYUV,这RGB色彩空间显示出了更好的效果。这些图像分割过程融合了边缘检测方法来产生更好的结果。基于近似推理或模糊推理的分割产生了可喜的成果。Huntsberger定义颜色的边缘为每个像素成员函数值的差异为零。通过使用了C迭代分割算法得到了C迭代算法由于本身性质来说是耗时的。Lim自动化的从粗到细的分割方法。这种方法基于阈值直方图和C迭代算法。Lambert和,缘流技术,基于模型的随机分割法。Markov模型的色彩分割应用也被研究。最后,Boyokov等人基于图像分割技术原理提出了一种颜色-纹理分割方法,把它看做一个加的架构和硬件实现。Perez和Koch提出了一种简化的适合执行在模拟集成电的色相描述。他们首次设计、制造并模拟了CMOS超大规模集成电路来计算标准化色相和色调。StichlingKleinjohann提出用硬件实现颜色分割算法,通过使用区域增长和合并法在飞利浦Trimedia微控制器上实现。这种系统每秒可处理25帧小图像,使用HW-SW系统。Leclerq和Braun在一个32位的摩托罗拉的控制器上实现了颜色分割80*600.02秒内识别备在索尼AIBO机器人身上实现。Johnston等人做了一个系统,使用了FPGA,实现了颜色分割和对象并提供实时处理。Koo等人做出了一个系统,用来分析磁图4FPGA的计算机上实现,并达到了五倍的加速比。Dillinger等人建立了一个基于FPGA的程序,实现了三维的分的图像上220个对象。止图像或。例如,一个ATR系统包括许多算法的组合,启发式分割,平缓,边缘法,特别是平滑、分割和边缘检测,在ATR系统中软件完成所需要的时间里,它们占FRS方法(平滑,边缘检这一部分的信息可以被随后的ATR系统执行对象在图像中的特征提取这一步骤所利以产生一个嵌入式的子系统。如本文所示,会看到一个嵌入式处理器,紧密耦合理器和可重配置的部分之间传送,将对这项工作的深入提出了。Stretch公司已经开发出S5000和S6000系列软件配置处理器,这都是基于Tensilica的RISC处C/C++平台上开发,包括本2FSR3节描述新的体5节会对本工作做出FRS1RGB工作的很好。(见第1节。图 FRS算法的数据流模糊算子,其中为每个相邻像素的领域都设置了相应的算子,如图2所示。图 333*318,以便带入下面的两个式子计我们实施的块的大小为3*3,这样对图像可以达到高度平滑的效果。每一个领域块的平均颜色都按照下面所示的函数(1)来进行计算。要做到平滑,必须测量中心像素和所有周围像素域之间的颜色对比度。像素(i,j)和领域块b之间的颜色对比,从原始图像RGB数值中用这些参数通过几何学来表示为下面的式子:4图 数。这些参数是从原始图像的RGB值由以下算得的:在该算法的第一个步骤中,h,si3*3(图3。一个物体,无论是色调,亮度,或者是阴影的差异,都具有相同的色相贯穿5图 意义上近邻近的区域:近空间域(物理近;或近集群域的颜色立方体(几乎个过滤器用作三大小分别是3*3,5*5,7*7的块,最小的一个被最先应用。ATR算法的前三个步骤,其中还包括三个子系统,每一个都为一个算法的主要方面来66CStretchS5000StretchC编写的语言,映射在可重构的部分,就是处理器上所谓的ISEF。每个使用可重构开发ISEF上的映射并使用各自的输入数据运行它的代码。并行代码完成后的结果返回到程序,串行代码继续执行。实现模块之间的通信将通过具有内部功能的S5000处理器。最初,彩片像素的RGB值在处理器的内部中,平滑模块通过这平滑子系统需要原始像素的RGB值作为输入和平滑后的图像数值作为输出。该平滑算法分为七个小步骤,每一个都被设计为一个单独的模块。结构如图7所示。7S5000处理器上并行工83*33所示。这些C(i,j),b7所示。S5000处理器可重构部分的限制之一是,用来在可重构的部分和处理器的软件部分之间进行通信的I/O寄存器(128位*3=384位)I/O寄存器的限制每次都会引起相邻两个窗口之间的颜色计算,两个相邻的3*3窗口包含了15个不同的像素(有一个的3像素,这就意味着15*8*3(像素的颜RGB模式)=360位<384位。这对算子块对C(i,j),b值的计算如图3所示。C(i,j),b的计算模块在可重构C(i,j),b的值包含乘法,这是对软件来C(i,j),b子窗口仅需要三个硬件时在这一阶段中,一个新的可重构的逻辑模块被开发。该模块把8个邻域子窗口的C(i,j),b的值,处理的像素的颜色和它们之间的差异计算值作为输入,完成这个计算后,7所示,第二个是将要被处理的子窗口的ID号。这个过程中每个计算需要9个时钟间的差异,如图7所示。计算的过程是一个繁重的问题,因此它发生在S5000中可像素*3RGB=147*8=1176位)。为了处理这个问题,我们考虑到一个有利条件就是,四C(i,j),b的在第一步的平滑算法中已经被可重构的硬件计算过了。在这3*37*7的窗口的像素值相加。最后的结果被7*7窗口的和所相除。C(i,j),b的值和四个窗口的方向(东南西北C(i,j),b的RGB*4个窗口=12*8=96位)作为输入,这887*7窗口的像素值作为输入。用颜色平滑单Cbs3*3的子窗口的颜色差异。如图3所示的四个方向的每一个方向的颜色差异,被暂时起来,被用于色调,饱和度和强度的每一个方向的计算。最后得出的值被当做色调差异模块的输入值在表S5000处理器的可重构从原始图像中产生了。这个信息被在S5000处理器的器中,并且作为下3699颜色分割子系统把S5000处理器器中的边缘检测像素值作为输入,并且把用不同颜色标注了对象的,这个可以用于下一步的图像处理系统,比如 Stretch处理器上的可重构部分来执行。它的输入包括处理后的像素值和周围有3*3窗口的像素值,并且给出信号值显示该像素是否可以被用作区集中起来,根据这个处理的信息把每一个区域添加一个ID。最后,区域平均颜色的计Stretch技术中可能会有限RGB模式上的级数。另外,软件乘法器和加法器被设计用于计算颜色反射值和每一个被扩展后的区域的颜色。最后的扩展算法FRSStretchFRSSW/HWStretch处理器上的实现。这个算法的S5000FPGA结构上建立,另外这个算法的其他部分被串行的C代码在嵌入式的Stretch处理器上执行。这个小节描述了这个工作中最重要最的部分,就是Stretch可重构处理器能用于实现FSR算法的方如上所述,像素或子窗口程序,和许多算法的不同阶段。这些都导致了FPGA上的可度简化成了定点计算,更快地使可重构模块执行。最后,FPGA和处理器之间的数据通ContentslistsContentslistsavailableatSciVerseMicroprocessorsandjournal/locate/micproMicroprocessorsandMicrosystems36(2012)Anembeddedsoftware-reconfigurablecolorsegmentationforimageprocessingTechnicalUniversityofCrete,ECEDept.,Chania,Crete,WrightStateUniversity,Engr.CollegeATRCenter,Dayton,OH45435,articleinfArticleAvailableonline17December:ReconfigurablearchitecturesImagesegmentationEmbeddedsystems
abstracImagesegmentationisoneofthefirstimportantanddifficultstepsofimage ysisandcomputervisionanditisconsideredasoneoftheoldestproblemsinmachinevision.Lay,severalsegmentationalgorithmshavebeendevelopedwithfeaturesrelatedtothresholding,edgelocationandregiongrowingtoofferanopportunityforthedevelopmentoffasterimage/ ysisandrecognitionsystems.Inaddition,fuzzybasedsegmentationalgorithmshaveessentiallycontributedtosynthesisofregionsforbetterrepresentationofobjects.Thesealgorithmshaveminordifferencesintheirperformanceandtheyallperformwell.Thus,theselectionofonealgorithmvs.anotherwillbebasedonsubjectivecriteria,or,drivenbytheapplicationitself.Here,alowcostembeddedreconfigurablearchitecturefortheFuzzylikereasoningsegmentation(FRS)methodispresented.TheFRSmethodhasthreestages(smoothing,edgedetectionandtheactualsegmentation).Theinitialsmoothingoperationisintendedtoremovenoise.Thesmootherandedgedetectoralgorithmsarealsoincludedinthisprocessingstep.Thesegmentationalgorithmusesedgeinformationandthesmoothedimagetofindsegmentspresentwithintheimage.InthisworktheFRSsegmentationalgorithmwasselectedduetoitsprovengoodperformanceonavarietyofapplications(facedetection,motiondetection,AutomaticTargetRecognition(ATR))andhasbeendevelopedinalowcost,reconfigurablecomputingtform,aimingatlowcostapplications.Inparticular,thispaperpresentstheimplementationofthesmoothing,edgedetectionandcolorsegmentationalgorithmsusingStretchS5000processorsandcomparesthemwithasoftwareimplementationusingthe .Thenewarchitectureispresentedindetailinthiswork,togetherwithresultsfromstandardben arksandcomparisonstoalternative.Thisisthefirstsuchimplementationthatweknowof,havingatthesametimehighthroughput,excellentperformance(atleastinstandardben arks)andlowcost.Ó2011ElsevierB.V.All Manycomputervision,patternrecognition,image ysisandobjectextractionsystemshavebeendevelopedduringthelastthirtyyears.Atthesametime,fuzzyandsemifuzzyclusteringalgorithmshavebeenalsopresentedfortheextractionandrecognitionofanobject’sfeatures.Inorderforthesesystemsandalgorithmstobesuccessfultheygenerallyhavetostartwitharobustsmoothingand/orsegmentationtechnique.Thus,imagesegmentationisanimportantstartingstepforalmostallvisionandpatternrecognitionmethodologies.Severalstudieshavebeendonetocategorizesegmentationintoclassesbasedoncharacteristics,suchasthresholdingorclustering,edgedetection,regiongrowing/merging*Correspondingnikolaos.bourbakis@(N.Bourbakis).
andothers[13].Inparticular,LeeandChung[4]showedthatthresholdingwouldusuallyproducegoodresultsinbimodalimagesonly,wheretheimagescompriseofonlyoneobjectanditsbackground.However,whentheobjectareaissmallcomparedtothebackgroundarea,orwhenboththeobjectandbackgroundhaveabroadrangeofgraylevels,selectingagoodthresholdisdifficult.Anotherweaknessofthistechniqueoccurswhenmultipleobjectsarepresentwithintheimage.Insuchcases,findingsharpvalleyswithinthehistogramisfurthercomplicated,andsegmentationresultsmaybeverypoor.Edgedetectionisanotherapproachassociatedtoimagesegmentation[5].Anedgeisdefinedasalocationwhereasharpchangeingraylevelorcolorisdetected.However,inthismethoditisdifficulttomaintainthecontinuityofthedetectededges;asegmentmustalwaysbeenclosedbyacontinuousedge.Regiongrowingormergingisathirdapproachforimagesegmentation[6].Inthiscase,large,easytofindcontinuousregionsorsegmentsaredetectedfirst.Afterwards,smallregionsmaybemergedbyusinghomogeneitycriteria[7,8].Onedisadvantageofregiongrowingandmergingistheinherentlysequential0141-9331/$-seefrontmatterÓ2011ElsevierB.V.Allrights G.Chrysosetal./MicroprocessorsandMicrosystems36(2012)natureofthisapproach.Often,theregionsproduceddependupontheorderinwhichthoseregionsgrowormerge.ColorsegmentationTheli turereportsdifferentapproachesforcolorsegmentation.Animportantcolorsegmentationmethodisthedevelopmentofdichromaticreflectionmodel[15,16],whichdescribesthecolorofreflectedlightasalinearcombinationofthecolorofsurfacereflection(highlights)andbodyreflection(objectcolor).Useofthismodeltotheregiongrowingandmergingmethod[6,17]producedimpressiveresults.Inthismethod,highlightedareasweremergedwiththematteareasofanobject.However,usinghardthresholdsthroughoutdegradedtheperformanceofthistechniquewithinitsintermediatestages.Therearesegmentationmethods[18,19]whichdonotsegmentthecolorimageintheRGBcolorspace,asitdoesnotcloselymodelthepsychologicalunderstandingofcolor.Insteadof,theychooseothercolorspaces,likeHISorYUV,whichproducebetterresultsthantheRGBcolorspace.Someoftheseimagesegmentationprocesseswerefusedwiththeedgelocationmethodtoproducebetterresults[20,21].Segmentationbasedonthetheoryofapproximatereasoningorfuzzylikereasoningproducedpromisingresults[22,23].Huntsberger[5]definedcoloredgesasthezerocrossingofdifferencesbetweenthemembershipvaluesofeachpixel.Thefuzzymembershipvaluesaregeneratedbyusingani tivecmeansegmentationalgorithmalthoughitistimeconsumingduetoitsi tivenature.Lim[24]presentedanautomatedcoarsetofinesegmentationmethod.Thisapproachisbasedonhistogramthresholdsforeachcolorandthecmeansalgorithm[25,26].Aninterestingapproach,proposedbyLambertandCarron[27],combinedthecolorspace(wherehuewasexplicitlydefinedandprocessedaccordingtoitsrelevancetochroma)andsymbolicrepresentationsandrulebasedsystems(usingcolorandluminancefeaturestodeterminehomogeneityamongpixels).Recently,moresegmentationtechniquesbasedoncolorandtexturehavebeenintroducedusingfeaturescommonlyobservedinmostimages,especiallyincolortexturedimagesofnaturalscenes.Extensiveresearchresultsonhumanperceptionofcolorandtexturearealsoavailableintheli ture,e.g.,uniformcolorspaces[64]orfilterbanks[3537].Forallthesereasons,mostsegmentationmethodsusecolorortextureaskeyfeaturesforimagesegmentation.Recently,severalattemptstocombinecolorandtexturehavebeenmadetoenhancethebasicperformanceofcolorortexturesegmentation.Theseattempts,namelycolortexturesegmentation,includeregiongrowingapproaches[3840],watershedtechniques[41],edgeflowtechniques[42],andstochasticmodelbasedapproaches[43,44].TheapplicationofMarkovmodelsoncolorsegmentationhasalsobeenstudied[45,46].LastlytheBoyokovet.al.[4749]approachtocolortexturesegmentationisbasedongraphcuttechniqueswhichfindanoptimalcolortexturesegmentationofacolortexturedimagebyregardingitasaminimumcutprobleminaweightedgraph.Therearemanyarchitecturesandhardwareimplementationsofcolorsegmentationalgorithmsinli ture.PerezandKoch[28]proposedtheuseofasimplifiedhuedescriptionsuitableforimplementationin ogVLSI.Theydesignedandfabricatedforthefirsttimean ogCMOSVLSIcircuitthatcomputesnormalizedcolorandhue.StichlingandKleinjohann[29]presentahardwareimplementationofcolorsegmentationalgorithmusingregiongrowingandmergingmethodsimplementedonaPhilipsTrimediamicrocontroller.Thesystemprocesses25framespersecrateforsmallimagesandusingaHWSWsystem.LeclerqandBraunl[30]implementedacolorsegmentationalgorithmona32bitMotorolacontrollerfor8060images.ThesystemwasusedfortheRobocupcompetitionandidentifiessmallobjectsinabout0.02s.Saffiotti
[31]presentstheimplementationofaseededregiongrowingsegmentationalgorithmonaSonyAIBOrobotusingthespecificdeviceCDTthatusesthresholdtechnique.Johnstonetal.[32]presentasystemthatimplementscolorsegmentationandobjecttrackingusinganFPGA(SpartanII)andofferingrealtimeprocessing.Kooetal.[33]presentasystemthatyzesmagneticresonanceimages.Thesystemwasimplementedonahighperformancereconfigurablecomputerusing4FPGAsandachievesa5speedupofthealgorithm.Dillingeretal.[34]builtanFPGAbasedcoprocessorwhichimplementsa3Dimagesegmentationachievinghighperformance.Yamaokaetal.[35]presentanovelalgorithmimplementedonanFPGAtrackingupto220objectson8060SegmentationforimageprocessingbasedImageprocessingsystemssuchasAutomaticTargetRecognition(ATR),FaceRecognition,andMotionDetection[14,5054,62]requirearobustandfastsegmentationalgorithm.Thus,thesesystemsuseaprocessforobjectoffeaturesextractionandrecognitionappliedtostillimagesand/or [913].Forinstance,anATRsystemconsistofacombinationofalgorithms,suchassmoothing,heuristicsegmentation,edgedetection,thinning,regiongrowing,fractals,etc.,appropriayselectedtorecognizetargetsundervariousconditions.Thesealgorithms,especiallythesmoothing,segmentationandedgedetectionconsumeasignificantamountofcomputingtimeneededforthesoftwarecompletioninanATRsystem.Colorsegmentationisamuchstudiedproblem[45,57,58],asitisusedinapplicationssuchasfacerecognition[55,56].Thus,thecontributionofthisworkisanarchitectureanddetailedhardwaredesignfortheimplementationofthethreetimeconsumingpartsoftheFRSmethodology(smoothing,edgedetectionandcolorsegmentation)[7,8,22,23,36],whichweredevelopedasindependentinhardwareas‘‘blackboxes’’toperformaspecificprocedure.Thefinalresultisanimagedividedintoitsobjectswhicharecoloredwiththesamecolor.ThispieceofinformationcanbeusedbythesubsequentstepsoftheATRsystemtoperformfeatureextractionoftheobjectsintheimage.ThecompletesystemwasfullydesignedinareconfigurableprocessorusingthetechnologyofStretch,Inc.Thisisalowcosttechnologywhichleadstoaneasilyembeddablesubsystem.Aswillbeshowninthispaper,thetightcouplingofanembeddedprocessorwithreconfigurablefabricallowsforanefficientimplementationofthealgorithm,however,thevastamountsofdatathatneedtobetransferredbetweenthememory,theprocessor,andthereconfigurablepartposechallengeswhichwillbepresentedindepthinthiswork.The [51]hasdevelopedtheseriesofS5000S6000softwareconfigurableprocessors,whichisbasedontheTensilicacoreRISCprocessorwithasmallembeddedreconfigurablepart.ThedesignflowcomprisesofsystemdevelopmentinC/C++,profilingofthecode,andmapitscriticalsectionstothereconfigurablefabricasspecial,hardwareimplementedinstructions.TheC/C++languageisusedtoprogramtheS5000processors.StretchCisaClikelanguagewhichincludessomeextensionsforhardwareimplementation.StretchCistheprogramminglanguagewhichmapsthecriticalpartsofthedesigninthereconfigurablepartsoftheprocessor.Therestofthispaperisorganizedasfollows:Section2describestheFSRsegmentationmethodologythatwasimplemented.Section3describesthenewarchitecture,itsmajorsubsystems,theirinterconnection,anditsmapontheStretchtechnology.Section4hasperformanceresultsandadetailedcomparisontopreviouslypublishedimplementations.Finally,Section5hassomeconclusionsfromthiswork.G.Chrysosetal./MicroprocessorsandMicrosystems36(2012) TheFRSsegmentationSegmentationisaprocessusedtofacilitatetheextractionofobjectsthatformanimage.TheFRSmethodology,whichisstudiedinthispaper,consistsofthreesteps(priortotherecognitionitself):smoothing,edgedetectionandcolorsegmentation.ThedataflowofsegmentationprocessisdescribedinFig.1.Inthiswork,aswillbeshownbelow,theHIS(hue,intensity,saturation)modelisused,fromoriginalRGBimages,anapproachwhichisquitetypicalandhasbeenshowninli ture(seeSection1)toworkwell.SmoothingTheimagescontainnoiseintroducedeitherbythecameraorbecauseoftheimage’stransmissionoveranoisymedium.Ineithercase,thenoisemustberemovedbeforeanyfurtherimageprocessingisapplied.Themostcommonwayofnoiseremovalistheuseoffilters.Animportantconceptforasmoothingalgorithmistheneighborhoodbetweentwopixels.Thisalgorithmallowsforafuzzydegreeofneighborhood,inwhichforeachneighboringpixelthereisthecorrespondingdegreeofneighborhood,asshowninFig.2.Eachpixel’scoloriscomparedwiththecolorofeachofitsneighboringblocks,asshowninFig.3.Thesizeofblocksforourimplementationwas33,whichresultstoastrongsmoothingoftheimage.TheaveragecolorforeachoftheneighboringblockswascalculatedtakingintoaccounttheneighborhoodmembershipfunctionasshownintheEq.(1).Forsmoothing,thecolorcontrastbetweenthecenterpixelandallofthesurroundingblocksmustbemeasured.Thecolorcontrastbetweenthepixel(i,j)andtheblockbistheEuclideandistanceintheRGB asshowninthefol
Fig.2.TableofneighborhooddegreeEdgeEdgedetectionistheprocessofthelimitspecificationoftheobjectsanimageconsistsof.Hue,IntensityandSaturation(representedash,i,andsrespectively)areonesetofparametersthatareusedtoevaluatepixels’edgestrengthwithinimages.TheseparametersarecomputedfromtheoriginalimageRGBvaluesbytheequationsbelow: 0:49rþ0:31gþ0:2b; 0:177rþ0:812gþ 0:01gþPlowingP
kþ3
116
xX X0
y
qPP
kþ3
yqpsk
z
RÞ2þ GÞ2þ B
ThestepsofthesmoothingalgorithmthatwereimplementedinthisworkareshowninFig.4andtheyarepresentedyticallyin[23].
l;
a2þb2; tan1aOriginal
EdgeFig.1.ThedataflowofFSR G.Chrysosetal./MicroprocessorsandMicrosystems36(2012)Fig.3.Eightneighboringblocksofsize33andfouredgedirections.Blocksarenumbered1–8suchthattheymaybereferredtoEqs.(1)and(2)(variableComputecolorpixelandit’s8neighboringblocksFOREACHPIXELINTHEIMAGEmincolorMaxcontrast<Mincontrast<FindthecolorofthesidewiththelowestcontrastmeasureComputeaveragecolorcontrastbetweenthenorthblockandtheotherthreeblocksplusthecontrastwiththepixelComputeaveragecolorcontrastbetweenthewestblockandtheotherthreeblocksplusthecontrastwiththepixelComputeaveragecolorcontrastbetweenthesouthblockandtheotherthreeblocksplusthecontrastwiththepixelComputeaveragecolorcontrastbetweentheeastblockandtheotherthreeblocksplusthecontrastwiththepixelcoloraroundthepixelwithablocksizeof7x7ComputeaveragebetweenthepixelandtheblockwiththeminimumcontrastReceFig.4.Theflowchartofsmoothingalgorithmforpixeli,jandablocksizeof3Inthefirststepsofthealgorithm,thevaluesoftheh,sandiarecomputedforalleight33blocksaroundapixel(Fig.3).Anobjecthasthesamehuethroughout,regardlessofvariancesinshades,high
tionsandintensities,thereforethehueshouldbenormalized.Thethreevalues(hue,intensityandsaturation)leadtothecalculationG.Chrysosetal./MicroprocessorsandMicrosystems36(2012) ofthepixelsthatareobjectedgesaccordingtothealgorithmpresentedin[23].TheflowdiagramoftheedgedetectionalgorithmisshowninFig.5.ColorsegmentationThesegmentationalgorithmusesedgeinformationandtheinformationofthesmoothedimagetofindsegments.Thestepsinvolvedinthissegmentationprocedurefollow:FindbigandcrispExpandsegmentsbasedonhomogeneityExpandsegmentsbasedonthedichromaticreflectionExpandsegmentsbasedonthedegreeoffarnessApplyan tiveThefirststepofthecolorsegmentationalgorithmistheoffindingbigandcrispsegments.Onceedgedetectionhas
performedonanimage,crispsegmentsaresurroundedbyedgepixelsortheimageboundary.Crispsegmentscanbedefinedasasetofpixelscompleysurroundedbyedgepixelsbelongingtoonlyoneobject.Thenextstepofthesegmentationalgorithmistheexpansionofthesegmentsbasedonspecificcriteriaofhomogeny.Theinitialimageisscannedandusingtheinformationthatresultedfromtheedgedetectiontheexistingsegmentsareexpandedbyaddingpixelswithhighsimilaritytothoseoftheexistingsegments.Thethirdstepofthecolorsegmentationprocessisthesegmentexpansionbasedonthedichromaticreflectionmodel.Usingthedichromaticreflectionmodel,someadjacentpixelsmaybemergedwiththepreviouslygrowingmattesegmentsaccordingtoafuzzymeasuresuchasthecustomizeddistancebetweenthemergingpixelandaclusterneinthecolor Tofurtherexpandsegments,the‘‘degreeoffarness’’measureisused.Anunassignedpixelcanbeclose(notfar)toaneighboringsegmentintwosenses:closeinthespatial FindthresholdedumsinthedirectionstoredforeachpixelCalculatethesaturation,intensityandhuecontrastforthe8neighboringForeachoneofCalculatethesaturation,intensityandhuecontrastforthe8neighboringForeachoneof4edgedirectionscalculatetheaveragesaturations,intensitiesandhuecontastsForeachoneof4edgedirectionscalculateμsandμiusingthelowmembershipfunctionCalculatetheaverageofthenormalizedhuecontrastandthefourhuecontrastforeachedgedirectiontl>Maxedgecandidacy>andfindCalculatethefouredgecandidacymeasureshueCalculateEvaluatefourbyoneMergethefouredgestrength G.Chrysosetal./MicroprocessorsandMicrosystems36(2012)close);orcloseinthecluster ofthecolorcube(almostofthesamecolor).Thedegreeoffarnessofapixeltoaneighboringsegmentisdefinedastheproductofthesetwomeasures.Specifically,thedegreeoffarnessforanygivenpixelthatwasusedistheabsolutecolorcontrastmultipliedbythegeometricdistance(inpixels)betweenthegivenpixelandthesegmentborder.Finally,afterthesegmentexpansioniscomplete,theresultingsegments’edgesaresmoothedusinganitivefilter.Thisfilterisusedforthreeblocksizesof33,55and77,withthesmallestonebeingappliedfirst.Inthefinalimageallthepixelsaregroupedinsegments,accordingtothedescribedcriteria.Eachsegmentoftheimageisconsideredtobeanindependentobjectandiscoloredwiththesamecolor.Thecolorofeachimageiscalculatedastheaveragecolorofthepixelsthatbelongtotheimage.Thealgorithmofcolorsegmentationispresentedindetailin[8].Anarchitectureforsmoothing,edgedetectionandThissectiondescribesthearchitecturesofthesmoothing,edgedetection,andcolorsegmentationalgorithmsthatweredevelopedusingtheStretchtechnology.Thesystem,whichimplementsthefirstthreestepsoftheATRalgorithm,consistsofthreesubsystems,oneforeachofthemainaspectsofthealgorithm.ThesystemblockdiagramispresentedinFig.6,andthearchitectureofeachsubsystemwillbedescribedinthefollowingsectionsofthispaper.EachsubsystemconsistsofserialCcodewhichrunsontheStretchS5000processorand‘‘hardware’’functions,writteninStretchClanguage,whicharemappedonthereconfigurablepart,thesocalledISEF,oftheprocessor.Eachreconfigurablefunctionisdevelopedusingthereconfigurableresourcesoftheprocessoranditscodeisexecutedinparallel.Eachtimetheserialprogramcallsa‘‘hardware’’function,theprocessorloadsitshardwaremapontheISEFandrunsitscodeusingtherespectiveinputdata.Aftertheparallelcodehasfinishedtheresultsreturntotheprocessorandtheserialcodeexecutioncontinues.ThecommunicationbetweentheimplementedcomponentstakescethroughtheinternalmemoryoftheS5000processor.Initially,thevaluesofcolorofthepictures’pixelsinRGBarestoredintheinternalmemoryoftheprocessor.Thesmoothingcomponentreadsthesevalueswhichareprocessedandtheyarestoredagaininthememoryofthesystem.Theprocesscontinuesfortheothertwosystem’scomponentswhichreadtheinputdatafromthememoryandstoretheprocesseddata.Finally,thesystemoutputsthesegmentedimagewherethedetectedobjectsofthepicturearecoloredwiththesamecolorinRGB.SmoothingsubsystemThesmoothingsubsystemtakesasinputtheinitialvaluesofpixelsintheRGBsystemandoutputsthevaluesofthesmoothedimage.Thesmoothingalgorithmisdividedintosevensmallsteps,
eachofwhichwasdesignedasaseparatecomponentasshowninblockdiagraminFig.7.ThedarkcomponentsoftheblockdiagramhavebeendevelopedinthereconfigurablepartofS5000processorandtheycanworkinparallelinordertoreducethetotalexecutiontimeofthesmoothingalgorithm.Initially,accordingtothesmoothingalgorithm,thereare8neighboring33windowsofeachpixel,asshowninFig.3.Thecolorofthesewindows,C(i,j),bvalues,areusedtofindthecolorcontrastbetweentheprocessingpixelandthesurroundwindows,accordingtoEq.(2).ThecalculationoftheC(i,j),btakesceinreconfigurablepartoftheprocessor,asshowninFig.7.OneoftherestrictionsofthereconfigurablepartofS5000processorsisthesmallnumberofI/Oregisters(128bits3=384bitsatmost)whichareusedforthecommunicationbetweenthereconfigurableandthesoftwarepartoftheprocessor.TheI/Orestrictionledtothecalculationofthecolorfortwoneighboringsubwindowseachtime,as2neighboring33windowscontain15differentpixels(thereisanoverlapof3pixels)whichmeans158bits/pixel3(asthecolorofthepixelsisinRGBmodel)=360bits<384bits.ThepairsofthesubwindowsthatcalculatedthevaluesofC(i,j),b,asshowninFig.3,are53,82,46and17.Also,thedirectionofthesubwindowsisimportantforthecalculationofthewindows’color.Thisledtotheimplementationoftwodifferentcomponentsforthecalculationofthesubwindows,oneforthehorizontalandanotherfortheverticaldirection.ThedesignC(i,j),b’scalculationmoduleonreconfigurablefabricwasmandatedbytheobservationsthat:(i)ThecalculationoftheC(i,j),bvaluecontainsmultiplications,whichisa‘‘heavy’’taskforthesoftwareand(ii)thisimplementationneedsonlythreehardwareclockcyclesforeachpairoftheC(i,j),b’ssubwindows.Finally,itisimportanttomentionthatasthereconfigurablepartoftheprocessordoesnotsupportfloatingpointarithmetic,fixedpointarithmeticwasusedforthevaluesoftheweightedtable.Thenextstepofthesmoothingalgorithmisthecalculationofthecolorcontrastbetweeneachofneighboringwindowsandtheprocessedpixel,usingtheEq.(3).Inthisstage,anewcomponentofreconfigurablelogicwasdeveloped.ThiscomponenttakesasinputtheC(i,j),bvaluesofthe8neighboringsubwindowsandthecoloroftheprocessingpixelandcalculatesthecontrastbetweenthe
温馨提示
- 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
- 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
- 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
- 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
- 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
- 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
- 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。
最新文档
- 鲜花烤奶课程设计
- 自来水收费系统课程设计
- 补牙系统课程设计
- 2025年度艺术品代购代发市场推广协议4篇
- 铁路线路课程设计
- 年度数字视频切换台市场分析及竞争策略分析报告
- 年度工艺礼品加工设备市场分析及竞争策略分析报告
- 2024年央行金融政策和法律法规测试题及答案汇编
- 二零二五年驾校场地租赁与师资力量引进协议3篇
- 重卡汽配配件课程设计
- 微信小程序运营方案课件
- 抖音品牌视觉识别手册
- 陈皮水溶性总生物碱的升血压作用量-效关系及药动学研究
- 安全施工专项方案报审表
- 学习解读2022年新制定的《市场主体登记管理条例实施细则》PPT汇报演示
- 好氧废水系统调试、验收、运行、维护手册
- 中石化ERP系统操作手册
- 五年级上册口算+脱式计算+竖式计算+方程
- 气体管道安全管理规程
- 《眼科学》题库
- 交通灯控制系统设计论文
评论
0/150
提交评论