同济大学摄影测量学课件

1Photogrammetry摄影测量学关泽群同济大学测量与国土信息工程系Chapter5DigitalImageandFeatureExtractionContents5.1ImageSamplingandResampling5.2PointFeatureExtraction5.3LineFeatureExtraction4DIGITALIMAGERYAdigitalimageisaregulararrayofpixels,orpictureelements,andisdescribedintermsofitsgeometryanditsradiometry.数字影像是一个灰度矩阵g,矩阵中每个元素gi,j,是一个灰度值,对应着光学影像或实体的一个微小区域,称为像元素或像元或像素（Pixel=Pictureelement）。各像元素的灰度值gi,j,代表其影像经采样与量化了的“灰度级”g0,0g0,1……g0,n-1g1,0g1,1……g1,n-1……………gm-1,0gm-1,1……gm-1,n-1数字影像或数字化影像6ToObtainDigitalImageEachpixelofadigitalimageisadirectorindirectsampleofthecontinuousincidentlightdistribution.–Adigitalsensorsamplesincidentlightdirectly.(可以直接装在空间飞行器中的扫描式传感器、数码相机、以及加有特殊装置的工业摄像机直接获得)–Anexampleofindirectsamplingisscanningapieceoffilmandconvertingitintoadigitalimage.(也可利用影像数字化器（如扫描仪等）对摄取的像片经过采样、量化获得——地形测图中航测通常使用)7Pixelsmaybearrangedinanyregularpattern,butinpracticemostsamplinggridsaresquareorrectangular(一般为矩形格网).Hexagonalgrids(六角形格网)havesometheoreticaladvantagesbuthaveseldombeenimplemented.Theratiobetweenpixelwidthandpixelheightisknownastheaspectratio（纵横比）;asquarepixelhasanaspectratioofone.Pixel,IFOVandGSD8Digitalimagesareoftendescribedintermsoftheirpixelsize,butcanalsobespecifiedintermsoftheinstantaneousfieldofview(IFOV)瞬时视场,whichistheangledeterminedbythepixelsizeandthefocallength.AsshowninFig.,asthefocallengthdecreasesorthepixelsizeincreases,theIFOVincreases.Figure.Instantaneousfieldofview(IFOV)andgroundsampledistance(GSD)–AlargerIFOVmeansthatonepixelcoversalargerportionofthesceneor,equivalently,thattheimagewillhavelowerspatialresolution.9Figure.Instantaneousfieldofview(IFOV)andgroundsampledistance(GSD)Instantaneousfieldofview(IFOV)瞬时视场groundsampledistance(GSD)地面采样间隔(距离)Anothermeasurerelatedtopixelsizeisthegroundsampledistance(GSD),whichistheprojectionofthepixelsizeontothegroundplane(Fig.).Pixel,IFOVandGSD10Thisisoftenusederroneouslyasasynonymforresolution.–Infact,theresolutionofadigitalimageisdeterminedbyboththesensorgeometryandbyfactorsexternaltothesensor,suchasatmosphericconditions,platformmotion,etc.TheactualresolutionmaybeasgoodasindicatedbytheGSD,butinpracticalapplicationsitisoftenlower.Pixel,IFOVandGSD11IFOVandGSDCalculationAsatelliteimagingsystemhaspixelsthatare0.015ramsquareandafocallengthof1000mm.Whatis(theIFOV?WhatistheGSDatnadir,assuminganorbitalaltitudeof650km?SOLUTIONTheIFOVisradIFOV000015.010002/015.0tan21andtheGSDisGSD=IFOV(650,000)=9.75m1.DigitalImageSampling13Digitalimagessampleacontinuousanaloginputimageatdiscretepointsandassignadiscreteintensityvalueateachpoint.Howcanwedeterminewhetheralltheinformationintheoriginalimageispresentinitssampleddigitalformor,equivalently,whatspatialsamplingrateisrequiredtofullycapturetheinput?14Theanswertothisquestionreliesonsamplingtheory.–Samplingtheorystatesthataband-limitedsignalwithahighestfrequencyofωcanbeexactlyreconstructedwithasamplingrateof2ω.（当采样频率大于信号中最高频率的2倍时，采样之后的数字信号完整地保留了原始信号中的信息）–Intermsofwavelength,asignalwithaminimumwavelengthofλmustbesampledwithasamplespacingofnogreaterthanλ/2.ThetheoreticalminimumsamplingraterequiredisknownastheNyquistrate.15Whathappensifwedon'tsampleatahighenoughrate?–Thehigher-frequencycomponentsofthesignal,whichcorrespondtothefinedetailsintheimage,areincorrectlyinterpretedaslower-frequencycomponents.Theseerroneouscomponentsconfusetheoverallsignalreconstruction,aphenomenonknownasaliasing（混淆）.16Fordigitalimages,thelensandthepixelbothactasapertures,limitingthemaximumdetailvisibleintheimage.Formostaerialcameras,thelens(andfilm,ifascannerisused)usuallyhavemuchhigherresolutionthanthecommonlyusedpixelsizescanrecord,sothepixelsizedeterminestheeffectiveaperture.DigitalImageSamplingSampling对实际连续函数模型离散化的量测过程Samplingpoint被量测的“点”是小的区域----像素Samplinginterval矩形的长与宽通常称为像素的大小Samplingtheorem(One-dimensionalimage）dxexgfGfxj2)()(Band-limitedfunctionSamplingProcessingkkxkxxkgxkxxgxgxs)()()()()()(Samplinginterval--xlfx21fliscutofffrequency(Nyquistrate)当采样间隔能使在函数g(x)中存在的最高频率中每周期取有两个样本时，则根据采样数据可以完全恢复原函数g(x)DigitalImageExpression1,11,10,11,11,10,11,01,00,0nmmmnngggggggggg)1,1,0()1,1,0(00mjyjyynixixxFrequencydomainFouriertransform2.DigitalImageResampling23DIGITALIMAGERESAMPLINGWeoftenneedtoresampleadigitalimageintoadifferentgeometry,forexample,tocreateanorthophoto.Anytypeofgeometrictransformation,includingsimilarity,affine,projective,polynomial,etc.,canbeappliedtoadigitalimageinexactlythesamewayastoanytwo-dimensionalsetofcoordinates.ImagematchingReferenceimages(主图像)Matchingimage(辅图像)25However,oncewehavecalculatedthenewimagecoordinates,whichareoftenfractional,wemustdecidewhichgrayvaluetoassigntothem.Anewvaluemustbeinterpolatedfromthepixelsneighboringthecoordinates;anumberofmathematicaltechniquesareavailable.Imageresamplingtheorem当欲知不位于矩阵（采样）点上的原始函数g(x,y)的数值时就需进行内插，此时称为重采样（resampling）Non-samplingpointImageresamplingmethodkllllkxkxfxkxfxkgxfxfxkxxkgxg)(2)(2sin)(22sin)()()(DigitalimageSincsamplingfunctionlfx2128Threecommonresamplingmethod