数字图像处理ppt第6章图像分割

DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.Woods第10章图像分割DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.Woods图像分割实例：车牌定位DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.Woods主要内容10.1间断检测10.2边缘连接和边界检测10.3门限处理10.4基于区域的分割10.5基于形态学分水岭的分割（自学）DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.WoodsDigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.WoodsDigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.Woods原理1、基于灰度的不连续性。（区域之间）2、基于灰度的相似性。（区域内部）DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.WoodsDigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.WoodsDigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.WoodsDigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.Woods10.1间断检测1点检测（1）原理用空域的高通滤波器来检测孤立点。-1-1-1-18-1-1-1-1TRzwRiiii91-1-2-1-212-2-1-2-1模板系数之和为零，表示灰度级为常数的区域，模板响应为零DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.Woods（1）这个模板公式是测量中心点和它的相邻点之间加权的差值；（2）基本思想：如果一个孤立的点与它周围的点很不相同，则很容易被这类模板检测到。DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.Woods（2）实例-1-1-1-18-1-1-1-1DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.Woods限制：这类检测是基于单像素间断的，并且检测器模板的区域有均匀的背景，所以这个检测过程有专用性。DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.Woods2线检测（1）原理模板检测DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.Woods2线检测若R1，R2，R3，R4对应4个模板的响应，则：若|Ri||Rj|,则此点被认为与i方向的线更相关。DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.Woods（2）实例DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.Woods3边缘检测DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.Woods3边缘检测（1）基础A、两种边缘模型光学系统、采样和其它图像采集的不完善性，边缘是模糊的DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.WoodsB、一阶导数和二阶导数DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.WoodsDigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.WoodsC、噪声对一阶导数和二阶导数的影响噪声对一阶和二阶导数都有影响，尤其对二阶导数影响较大，因此，在检测边缘前应该考虑平滑处理。左上角的图像分割是无噪声的，第一列的其它三幅附加零均值且标准差为0.1，1.0和10.0的高斯噪声DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.Woods（2）梯度算子A、梯度算子)arctan(),()()()(21222122xyyxyxyxGGyxaGGGyfxfGGfmagyfxfGGf边缘在(x,y)处的方向与此点的梯度向量的方向垂直。DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.WoodsDigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.WoodsDigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.WoodsDigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.WoodsDigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.WoodsC、实例DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.WoodsDigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.Woods对角Sobel模板的响应DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.Woods（3）拉普拉斯算子A、拉普拉斯算子),(4)1,()1,(),1(),1(),(2)1,()1,(),(),(2),1(),1(),(2222222222yxfyxfyxfyxfyxffyxfyxfyxfyyxfyxfyxfyxfxyxfyfxffDigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.WoodsDigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.WoodsDigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.Woods•Considerthefunction:•TheLaplacianofhis•TheLaplacianofaGaussiansometimesiscalledtheMexicanhatfunction.ItalsocanbecomputedbysmoothingtheimagewiththeGaussiansmoothingmask,followedbyapplicationoftheLaplacianmask.deviationstandardthe:andwhere)(222222yxrerhr2224222)(rerrhTheLaplacianofaGaussian(LoG)AGaussianfunctionDigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.WoodsDigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.Woods零交叉求边缘正值为白色，负值为黑色标记白色和黑色区域过渡点空心粉效应DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.Woods比较图像(b）和(g），有以下结果：（1）零交叉点图像的边缘比梯度图像边缘细；（2）（g)中图像的边缘来自许多闭合环的零交叉点。这种空心粉效应是这种方法的缺陷。DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.Woods10.2边缘连接和边界检测如何将边缘像素组合成有意义的边缘？DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.Woods10.2边缘连接和边界检测1基于局部处理的边缘点连接分析图像中每个点(x,y)的小邻域（如3*3或5*5）内像素的特点，将满足相似性准则的点连接起来，形成边缘。基于梯度向量或梯度向量的方向。AyxayxaEyxfyxf),(),(),(),(0000梯度角度DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.Woods原始图像垂直Sobel算子水平Sobel算子梯度大于25，方向差小于15度所有点连接。小断裂连接和孤立短线清除DigitalImageProcessing,2nded.©2002R.C.Gonzalez&R.E.Woods2通过Hough变换进行整体处理（1）问题的提出在找出边界点集之后，需要连接，形成完整的边界图形描述。DigitalImageProcessing,2nded.