机器视觉课件.ppt

《机器视觉课件.ppt》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。

1、Course4.RegionsDef.ofRegion:Asetofconnectedimagepixels,withsimilarproperties.similargraylevelsimilarcolorsimilartexturesoon……RegionmaybeseparatedadjacentoverlappedincludedImportantprincipleofregionprocess:——homogeneitypredicate(valuesimilarity)——specialproximity1.RegionRepresentation1)Arrayreprese

2、ntation:GivenaregionRinimageS,ItsarrayrepresentationAis:—sameasbinaryimagerepresentation—oftenbeusedasamaskoverintensityimagetoselecttheregionforproperprocessing.2)PyramidsrepresentationPyramidrepresentationofaNxNimage(N=2k)containsklayersofreducedimage.Eachpixelinanupperlayerimagecorrespondsto4pi

3、xelsinthelowerlayerimage,i.e.,Pixelsvalueofupperlayerimagecanbeobtainedby——averagingthecorrespondingpixelsoflowerlayerimage,or——subsampling(e.g.chooseupper-leftpixelvalue)3)QuadTree(Quad=4)Children:ifaregionSispartitionedinto4subregion,inquadtree,eachsubregionnodeiscallachildofthenodeoftheoriginal

4、region.1234S1234Colorofnode:Algorithm:Setanimageasaregion,andrepresentsitasanodeinquadtree(0-level)Splittheregioninto4subregions,andrepresentthemin4childnodes.Checkthecolorofeachchildnode:——ifcolorisblackorwhite,setnodeas“leafnode”,nofurthersplitting.——ifcolorisgray,gotostepb).d)Stopifallnodesaree

5、itherblackorwhite.4)RegionAdjacencyGraph:——Emphasizetheadjacencyofregions——node:region——link:commonboundarybetweenregions.5)PictureTree:——Emphasizeregioninclusionwithinanotherregion.——Recursivelysplitanimageintoregions,andtheprocessstopswhennoregioncanbefurthersplitted(homogeneitypredicate)2.Regio

6、nSegmentationSupposeanimageAiswellsegmentedintonregionsRi,i=1,2,…,n,theremustbe:WhereP(.)ishomogeneitypredicatemeasurement.Thresholding:(1)P-TileThresholingIfapre-knowledgeofobjectsizeSisgiven,theareapercentageoftheobjectinimageAcanbegotbyP=S/A.Thus,wesetthevalueofthresholdatTsuchthatwhere0

7、canbeobtainedfromimage’shistograph.(2)Optimalthresholding:SupposeanimagecontainonlytwoprincipalbrightnessregionsandbothobeyGaussiandistributioninintensity.Darkregion:Brightregion:wesetpartitionthresholdatT,errorp