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
HighdynamicrangeimagingDigitalVisualEffectsYung-YuChuangwithslidesbyFredoDurand,BrianCurless,SteveSeitz,PaulDebevecandAlexeiEfrosCameraisanimperfectdeviceCameraisanimperfectdeviceformeasuringtheradiancedistributionofascenebecauseitcannotcapturethefullspectralcontentanddynamicrange.Limitationsinsensordesignpreventcamerasfromcapturingallinformationpassedbylens.CamerapipelinelenssensorshutterAssumeastaticscene,Thus,Lisnotafunctionoftime.Camerapipeline12bits8bitsReal-worldresponsefunctionsIngeneral,theresponsefunctionisnotprovidedbycameramakerswhoconsideritpartoftheirproprietaryproductdifferentiation.Inaddition,theyarebeyondthestandardgammacurves.Theworldishighdynamicrange11,50025,000400,0002,000,000,000TheworldishighdynamicrangeRealworlddynamicrangeEyecanadaptfrom~10-6to106cd/m2Often1:100,000inasceneTypical1:50,max1:500forpictures10-6106RealworldHighdynamicrangespotmeterShortexposure10-610610-6106RealworldradiancePictureintensitydynamicrangePixelvalue0to255Longexposure10-610610-6106RealworldradiancePictureintensitydynamicrangePixelvalue0to255CameraisnotaphotometerLimiteddynamicrangePerhapsusemultipleexposures?Unknown,nonlinearresponse
NotpossibletoconvertpixelvaluestoradianceSolution:Recoverresponsecurvefrommultipleexposures,thenreconstructtheradiancemapVaryingexposureWaystochangeexposureShutterspeedApertureNeutraldensityfiltersShutterspeedNote:shuttertimesusuallyobeyapowerseries–each“stop”isafactorof2¼,1/8,1/15,1/30,1/60,1/125,1/250,1/500,1/1000secUsuallyreallyis:¼,1/8,1/16,1/32,1/64,1/128,1/256,1/512,1/1024secVaryingshutterspeedsHDRIcapturingfrommultipleexposuresCaptureimageswithmultipleexposuresImagealignment(evenifyouusetripod,itissuggestedtorunalignment)ResponsecurverecoveryGhost/flareremovalImagealignmentWewillintroduceafastandeasy-to-implementmethodforthistask,calledMedianThresholdBitmap(MTB)alignmenttechnique.Consideronlyintegraltranslations.Itisenoughempirically.TheinputsareNgrayscaleimages.(YoucaneitherusethegreenchannelorconvertintograyscalebyY=(54R+183G+19B)/256)MTBisabinaryimageformedbythresholdingtheinputimageusingthemedianofintensities.WhyisMTBbetterthangradient?Edge-detectionfiltersaredependentonimageexposuresTakingthedifferenceoftwoedgebitmapswouldnotgiveagoodindicationofwheretheedgesaremisaligned.SearchfortheoptimaloffsetTryallpossibleoffsets.GradientdescentMultiscaletechniquelog(max_offset)levelsTry9possibilitiesforthetoplevelScaleby2whenpassingdown;tryits9neighborsThresholdnoiseignorepixelsthatareclosetothethresholdexclusionbitmapEfficiencyconsiderationsXORfortakingdifferenceANDwithexclusionmapsBitcountingbytablelookupResultsSuccessrate=84%.10%failureduetorotation.3%forexcessivemotionand3%fortoomuchhigh-frequencycontent.Recoveringresponsecurve12bits8bitsRecoveringresponsecurveWewanttoobtaintheinverseoftheresponsecurve0255Dt=
1/4secDt=
1secDt=
1/8secDt=
2secImageseriesDt=
1/2secRecoveringresponsecurve•1•1•1•1•1•3•3•3•3•3•2•2•2•2•20255Dt=
1/4secDt=
1secDt=
1/8secDt=
2secImageseriesDt=
1/2secRecoveringresponsecurve•1•1•1•1•1•3•3•3•3•3•2•2•2•2•2Xij=lnXijIdeabehindthemathln2IdeabehindthemathEachlineforascenepoint.TheoffsetisessentiallydeterminedbytheunknownEiIdeabehindthemathNotethatthereisashiftthatwecan’trecoverBasicideaDesignanobjectivefunctionOptimizeitMathforrecoveringresponsecurveRecoveringresponsecurveThesolutioncanbeonlyuptoascale,addaconstraintAddahatweightingfunctionRecoveringresponsecurveWewantIfP=11,N~25(typically50isused)Wepreferthatselectedpixelsarewelldistributedandsampledfromconstantregions.Theypickedpointsbyhand.ItisanoverdeterminedsystemoflinearequationsandcanbesolvedusingSVDHowtooptimize?1.SetpartialderivativestozeroHowtooptimize?Setpartialderivativestozero
SparselinearsystemAx=b256nn×p1254g(0)g(255)lnE1lnEn:::QuestionsWillg(127)=0alwaysbesatisfied?Whyorwhynot?Howtofindtheleast-squaresolutionforanover-determinedsystem?Least-squaresolutionforalinearsystemTheyareoftenmutuallyincompatible.Weinsteadfindxtominimizethenormoftheresidualvector.Iftherearemultiplesolutions,weprefertheonewiththeminimallength.Least-squaresolutionforalinearsystemIfweperformSVDonAandrewriteitasthenistheleast-squaresolution.pseudoinverseProofProofLibrariesforSVDMatlabGSLBoostLAPACKATLASMatlabcodeMatlabcodefunction[g,lE]=gsolve(Z,B,l,w)n=256;A=zeros(size(Z,1)*size(Z,2)+n+1,n+size(Z,1));b=zeros(size(A,1),1);k=1;%%Includethedata-fittingequationsfori=1:size(Z,1)forj=1:size(Z,2)wij=w(Z(i,j)+1);A(k,Z(i,j)+1)=wij;A(k,n+i)=-wij;b(k,1)=wij*B(i,j);k=k+1;endendA(k,129)=1;%%Fixthecurvebysettingitsmiddlevalueto0k=k+1;fori=1:n-2%%IncludethesmoothnessequationsA(k,i)=l*w(i+1);A(k,i+1)=-2*l*w(i+1);A(k,i+2)=l*w(i+1);k=k+1;endx=A\b;%%SolvethesystemusingSVDg=x(1:n);lE=x(n+1:size(x,1));RecoveredresponsefunctionConstructingHDRradiancemapcombinepixelstoreducenoiseandobtainamorereliableestimationReconstructedradiancemapWhatisthisfor?HumanperceptionVision/graphicsapplications
AutomaticghostremovalbeforeafterWeightedvarianceMovingobjectsandhigh-contrastedgesrenderhighvariance.RegionmaskingThresholding;dilation;identifyregions;BestexposureineachregionLensflareremovalbeforeafterEasierHDRreconstructionrawimage=12-bitCCDsnapshot
EasierHDRreconstructionXij=Ei*Δtj
Exposure(X)Δt12bytesperpixel,4foreachchannelsignexponentmantissaPF7685121<binaryimagedata>FloatingPointTIFFsimilarTextheadersimilartoJeffPoskanzer’s.ppm
imageformat:PortablefloatMap(.pfm)(145,215,87,149)=(145,215,87)*2^(149-128)=(1190000,1760000,713000)(145,215,87,103)=(145,215,87)*2^(103-128)=(0.00000432,0.00000641,0.00000259)Ward,Greg."RealPixels,"inGraphicsGemsIV,editedbyJamesArvo,AcademicPress,1994Radianceformat(.pic,.hdr,.rad)RedGreenBlueExponent32bits/pixelILM’sOpenEXR(.exr)6bytesperpixel,2foreachchannel,compressedsignexponentmantissa
Severallosslesscompressionoptions,2:1typicalCompatiblewiththe“half”datatypeinNVidia'sCgSupportednativelyonGeForceFXandQuadroFX
Availableat/RadiometricselfcalibrationAssumethatanyresponsefunctioncanbemodeledasahigh-orderpolynomialNoneedtoknowexposuretimeinadvance.UsefulforcheapcamerasZXMitsunagaandNayarTofindthecoefficientscmtominimizethefollowingAguessfortheratioof
MitsunagaandNayarAgain,wecanonlysolveuptoascale.Thus,addaconstraintf(1)=1.ItreducestoM-1variables.Howtosolveit?MitsunagaandNayarWesolvetheaboveiterativelyandupdatetheexposureratioaccordinglyHowtodetermineM?SolveuptoM=10andpickuptheonewiththeminimalerror.Noticethatyouprefertohavethesameorderforallchannels.Usethecombinederror.Robertsonet.al.Givenand,thegoalistofindbothandMaximumlikelihoodRobertsonet.al.repeatassumingisknown,optimizeforassumingisknown,optimizeforuntilconvergeRobertsonet.al.repeatassumingisknown,optimizeforassumingisknown,optimizeforuntilconvergeRobertsonet.al.repeatassumingisknown,optimizeforassumingisknown,optimizeforuntilconvergeRobertsonet.al.repeatassumingisknown,optimizeforassumingisknown,optimizeforuntilconvergenormalizesothatSpaceofresponsecurvesSpaceofresponsecurvesPatch-BasedHDRHDRVideoHighDynamicRangeVideo
SingBingKang,MatthewUyttendaele,SimonWinder,RichardSzeliskiSIGGRAPH2003
videoAssortedpixelAssortedpixelAssortedpixelAVersatileHDRVideoSystem
videoAVersatileHDRVideoSystemHDRbecomescommonpracticeManycamerashasbracketexposuremodesiPhone4hasHDRoption,butitismoreexposure
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
- 文明城市创建活动阶段性工作总结
- Linux系统中的AI应用
- 中暑健康知识题库
- 预防脱肛指导
- 医院消防线路安全规范
- 研究生初期就业前景分析
- 对(间、邻)二甲苯装置操作工安全生产能力模拟考核试卷含答案
- 两栖类养殖工标准化强化考核试卷含答案
- 再生物资加工处理工安全应急模拟考核试卷含答案
- 粉末冶金模具工岗中风险识别考核试卷含答案
- 湖北省黄冈市2026-2027学年八上数学期末教学质量检测模拟试题含解析
- GA/T 1482-2026机动车驾驶人安全教育内容和方法
- 2026年北京市中考英语试卷真题(含答案及解析)
- 2026 年注塑车间主管上半年述职报告
- 安徽九华山旅游发展股份有限公司招聘笔试真题2025
- Q2桥门式起重机考试题题库及答案
- 2026年安徽省淮北市辅警考试试题解析及答案
- 【新教材】人教版(2024)八年级下册物理期末检测试卷 3套(含答案)
- 2026上海博物馆公开招聘12名工作人员备考题库及参考答案详解
- 零星维修施工方案
- 2026年山东定期医师考核题库及答案
评论
0/150
提交评论