
机器视觉chapter08.ppt
30页视觉与智能系统实验室第八章第八章 纹纹 理理视觉与智能系统实验室Atypicaltexturedimage.Formaterialssuchasbrush,grass,foliageandwater,ourperceptionofwhatthematerialisisquiteintimatelyrelatedtothetexture.视觉与智能系统实验室Brick Brick Stone视觉与智能系统实验室视觉与智能系统实验室8.1 概 述(1) 纹理定义:• 纹理是指图像强度局部变化的重复模式• Texture tells us information about spatial arrangement of the colors or intensities in an image.• 描述性定义:描述性定义:Firstly,viewsoflargenumbersofsmallobjectsareFirstly,viewsoflargenumbersofsmallobjectsareoftenbestthoughtofastextures.oftenbestthoughtofastextures.Secondly,manysurfacesaremarkedwithorderlySecondly,manysurfacesaremarkedwithorderlypatternsthatlooklikelargenumbersofsmallobjects.patternsthatlooklikelargenumbersofsmallobjects.•纹理元(纹理元(texeltexel))视觉与智能系统实验室由地板砖构成的地板纹理示意图(a)远距离观察时的纹理图像;(b)近距离观察时的纹理图像Textureisaphenomenonthatiswidespread,easytoTextureisaphenomenonthatiswidespread,easytorecognizeandhardtodefine.recognizeandhardtodefine.(2)纹理尺度Typically,whetheraneffectisreferredtoastextureornotdependsonthescaleatwhichitisviewed.Aleafthatoccupiesmostofanimageisanobject,butthefoliageofatreeisatexture.(3) 纹理问题 对机器视觉来说,纹理是为了分割和识别场景或物体表面类型而产生的一种视觉标记. 纹理分析包含有三个主要问题:• 纹理分类 从给定的一组纹理集中识别给定的纹理区域。
•纹理分割 自动确定图像中各种纹理区域之间的边界• 从纹理恢复形状透视投影产生的纹理模式来确定物体的三维形状•Texturesynthesisseekstoconstructlargeregionsoftexturefromsmallexamplemages.Wedothisbyusingtheexampleimagestobuildprobabilitymodelsofthetexture,andthendrawingontheprobabilitymodeltoobtaintexturedimages.(4) 纹理分析算法分为两大类:• 统计分析纹理基元小/微纹理 Statistical analysis: Texture is a quantitative measure of the arrangement of intensities in a region.• 结构分析大纹理基元 Structural analysis : Texture is a set of primitive texels in some regular or repeated relationship. 一幅具有三个灰度级的图像(1) 灰度级同现矩阵(gray-level co-occurrence matrix)P[i,j] 一个二维相关矩阵: 规定一个位移矢量d=(dx,dy) 计算被d分开且具有灰度级i和j的所有像素对数 举例灰度级同现矩阵,距离向量为d(1,1)8.2 纹理分析统计方法纹理分析统计方法0 1 2 i(a)棋格图像(b)距离为d=(1,1)的灰度级同现矩阵(c)距离为d=(1,0)的灰度级同现矩阵(a) (b) (c)规范化同现矩阵N[i,j]用于测量灰度级分布随机性的一种特征参数叫做熵,定义为:注意:当矩阵的所有项皆为零时熵值最高.这样的矩阵对应的图像不存在任何规定位移向量的优先灰度级.用灰度级同现矩阵定义能量特征、对比度特征和均匀度特征:(2) 自相关法一幅图像的自相关(Auto-correlation)函数定义为:测量不同粗细纹理示意图视觉与智能系统实验室(3) 用于纹理测量的Law能量法使用局部模板来检测各类纹理,比如视觉与智能系统实验室8.3 纹理的结构分析纹理的结构分析• 纹理基元大. • 纹理的结构分析法分为三步:图像增强;基元提取;计算纹理基元的特征参数及构成纹理的结构参数.• 纹理基元特征参数及纹理基元参数包括基元的尺寸、偏心、矩量、位置和姿态等。
由等间距排列的圆点形成的纹理图(a)原始纹理图 (b)图像受到噪音的污染导致的随机线条视觉与智能系统实验室(1) 纹理基元的提取二值化方法视觉与智能系统实验室(2)ExtractingImageStructurewithFilterBanksThereisastrongresponsewhentheimagepatterninaneighbourhoodlookssimilartothefilterkernel,andaweakresponsewhenitdoesn’t.Asetofeightfiltersusedforexpandingimagesintoaseriesofresponses.Thesefiltersareshownatafixedscale,withzerorepresentedbyamid-greylevel,lightervaluesbeingpositiveanddarkervaluesbeingnegative.Theyrepresenttodistinctspots,andsixbars.视觉与智能系统实验室(a)SpotsandBarsbyWeightedSumsofGaussians•Aspot:givenbyaweightedsumofthreeconcentric,symmetricGaussians(withweights1,-2and1,andsigmas0.62,1and1.6).•Anotherspot:givenbyaweightedsumoftwoconcentric,symmetricGaussians,withweights1and-1,andcorrespondingsigmas0.71and1.14.•Aseriesoforientedbars,consistingofaweightedsumofthreeorientedGaussians,whichareoffsetwithrespecttooneanother.Therearesixversionsofthesebars;eachisarotatedversionofahorizontalbar.TheGaussiansinthehorizontalbarhaveweights-1,2,and-1.Theyhavedifferentsigmasinthexandintheydirections;thexvaluesareall2,andtheyvaluesareall1.Thecentersareoffsetalongtheyaxis,lyingat(0,1),(0,0)and(0,-1).视觉与智能系统实验室Generally,spotfiltersareusefulbecausetheyrespondstronglytosmallregionsthatdifferfromtheirneighbours(forexample,oneithersideofanedge,orataspot).Theotherattractionisthattheydetectnon-orientedstructure.Barfilters,ontheotherhand,areoriented,andtendtorespondtoorientedstructure(thispropertyissometimes,ratherloosely,describedasanalysingorientationorrepresentingorientation).视觉与智能系统实验室Atthetop,animageofabutterflyatafinescale,andbelow,theresultofapplyingeachofthefilterstothatimage.Theresultsareshownasabsolutevaluesoftheoutput,lighterpixelsrepresentingstrongerresponses,andtheimagesarelaidoutcorrespondingtothefilterpositioninthetoprow.视觉与智能系统实验室Theinputimageofabutterflyandresponsesofthefiltersatacoarserscale视觉与智能系统实验室(b)SpotsandBarsbyGaborFiltersThekernelslooklikeFourierbasiselementsthataremultipliedbyGaussians,meaningthataGaborfilterrespondsstronglyatpointsinanmagewheretherearecomponentsthatlocallyhaveaparticularspatialfrequencyandorientation.Gaborfilterscomeinpairs,oftenreferredtoasquadraturepairs;oneofthepairrecoverssymmetriccomponentsinaparticulardirecton,andtheotherrecoversantsymmetriccomponents.Themathematicalforms视觉与智能系统实验室视觉与智能系统实验室Thesefiltersareshownatafinerscale视觉与智能系统实验室8.4 纹理分割视觉与智能系统实验室视觉与智能系统实验室8.5 从纹理恢复形状从纹理恢复形状 纹理基元的尺寸、形状、和密度等变化为表面形状和姿态估计提供了依据.从纹理恢复形状算法正是利用这些纹理基元的变化特性,从二维图像恢复三维信息。
纹理平面的倾斜角度为α在d(0,y’)处的椭圆最小直径为:8.6 从纹理恢复形状 为研究图像平面特性的变化,如尺寸、形状、密度和纹理基元的形态可以用精确的方法去刻化图像平面中的每一个基元 对于含有噪声的更复杂的恢复级纹理,精确估计图像平面特征将很困难的。












