sift算法英文详解

《sift算法英文详解》由会员上传分享，免费在线阅读，更多相关内容在教育资源-天天文库。

SIFT:ScaleInvariantFeatureTransformThealgorithmSIFTisquiteaninvolvedalgorithm.Ithasalotgoingonandcanbecomeconfusing,SoI’vesplituptheentirealgorithmintomultipleparts.Here’sanoutlineofwhathappensinSIFT.ConstructingascalespaceThisistheinitialpreparation.Youcreateinternalrepresentationsoftheoriginalimagetoensurescaleinvariance.Thisisdonebygeneratinga“scalespace”.LoGApproximationTheLaplacianofGaussianisgreatforfindinginterestingpoints(orkeypoints)inanimage.Butit’scomputationallyexpensive.Sowecheatandapproximateitusingtherepresentationcreatedearlier.FindingkeypointsWiththesuperfastapproximation,wenowtrytofindkeypoints.ThesearemaximaandminimaintheDifferenceofGaussianimagewecalculateinstep2GetridofbadkeypointsEdgesandlowcontrastregionsarebadkeypoints.Eliminatingthesemakesthealgorithmefficientandrobust.AtechniquesimilartotheHarrisCornerDetectorisusedhere.AssigninganorientationtothekeypointsAnorientationiscalculatedforeachkeypoint.Anyfurthercalculationsaredonerelativetothisorientation.Thiseffectivelycancelsouttheeffectoforientation,makingitrotationinvariant.GenerateSIFTfeaturesFinally,withscaleandrotationinvarianceinplace,onemorerepresentationisgenerated.Thishelpsuniquelyidentifyfeatures.Letssayyouhave50,000features.Withthisrepresentation,youcaneasilyidentifythefeatureyou’relookingfor(say,aparticulareye,orasignboard).Thatwasanoverviewoftheentirealgorithm.Overthenextfewdays,I’llgothrougheachstepindetail.Finally,I’llshowyouhowtoimplementSIFTinOpenCV!WhatdoIdowithSIFTfeatures?Afteryourunthroughthealgorithm,you’llhaveSIFTfeaturesforyourimage.Onceyouhavethese,youcandowhateveryouwant.Trackimages,detectandidentifyobjects(whichcanbepartlyhiddenaswell),orwhateveryoucanthinkof.We’llgetintothislateraswell.Butthecatchis,thisalgorithmispatented.>..