二維ITK連通區域提取

• 說明
• • 主要函數介紹
  • 細節分析
  • • 檢測圖像
    • 背景色
    • 編號順序
  • 完整代碼

說明

本文介紹2維ITK圖像連通區域的提取，對如下圖所示的二維圖像，進行連通區域提取，并打上不同的標簽，染上不同的顏色。這里以按連通區域像素個數大小進行排序，保留前幾個最大的連通區域。

提取前：

提取著色后

.
.
.

主要函數介紹

itk::ConnectedComponentImageFilter<ImageType, OutputImageType>;
該filter值檢測連通區域，不編號，不排序。

itk::LabelShapeKeepNObjectsImageFilter;
該filter對連通區域進行排序，然后只保留最大的幾個連通區域。

itk::LabelImageToShapeLabelMapFilter< OutputImageType, LabelMapType> ;
為保留的每個連通區域打上標簽（標號），編號順序為找到不同區域的順序.

雖然LabelShapeKeepNObjectsImageFilter調用SetAttribute 之后，按照attribute對連通區域大小進行了排序，但該排序并不反映在label上。即大的label并不一定對應大的size。反映在 SetNumberOfObjects(n) 設定保留區域之后，只保留排序后前n個區域的label。

itk::RescaleIntensityImageFilter<OutputImageType, ImageType>；
上一步圖像只有標簽，圖像上沒有直接的區別，本步對保留下來的不同連通區域進行不同的染色，以示區分。

細節分析

檢測圖像

itk::ConnectedComponentImageFilter，
根據該filter的描述，只對二值圖進行檢測，零值為背景色，非零值為前景色，即待檢測物體。

背景色

labelShapeKeepNObjectsImageFilter->SetBackgroundValue(0);
該函數設置背景色，只影響輸出的圖像背景色。此處黑色像素視為背景色。
使用該函數需要注意，默認值為該像素值類型的非正最小值NumericTraits::NonpositiveMin()，自定義使用其他值可能造成結果非預期（原因尙未探知，知者望告知），一般設為默認即可。

編號順序

對上圖進行檢測，保留5個連通區域，從大到小排序應該是ABCDE，但是編號順序是按照圖像像素遍歷順序進行編號，此處行優先進行遍歷，找到的連通區域的編號1->A, 2->B，3->D，4->C，5->E。反映在著色順序上，按照12345的順序由深色到白色進行著色。
注意著色時
rescaleFilter->SetOutputMinimum(0);
rescaleFilter->SetOutputMaximum(255);
設置著色的范圍，范圍設置太廣，可能造成部分標簽染色與背景色太接近而看不清。

完整代碼

#include "itkImage.h" #include "itkImageFileReader.h" #include "itkConnectedComponentImageFilter.h" #include "itkLabelShapeKeepNObjectsImageFilter.h" #include "itkRescaleIntensityImageFilter.h" #include "itksys/SystemTools.hxx" #include <sstream> #include "itkPNGImageIOFactory.h" #include "itkLabelImageToShapeLabelMapFilter.h" #include "itkSimilarity3DTransform.h" #include "itkResampleImageFilter.h" # define ENABLE_QUICKVIEW 1#ifdef ENABLE_QUICKVIEW # include "QuickView.h" #endifint main(int argc, char* argv[]) {itk::PNGImageIOFactory::RegisterOneFactory();constexpr unsigned int Dimension = 2;using PixelType = unsigned short;using ImageType = itk::Image<PixelType, Dimension>;using ReaderType = itk::ImageFileReader<ImageType>;ReaderType::Pointer reader = ReaderType::New();std::string fileName = R"(E:\Yinyang30rotate.png)";reader->SetFileName(fileName);try{reader->Update();}catch (const std::exception& e){std::cout << e.what() << std::endl;}using OutputImageType = itk::Image<unsigned short, Dimension>;using ConnectedComponentImageFilterType = itk::ConnectedComponentImageFilter<ImageType, OutputImageType>;ConnectedComponentImageFilterType::Pointer connected = ConnectedComponentImageFilterType::New();connected->SetInput(reader->GetOutput());connected->Update();std::cout << "Number of objects: " << connected->GetObjectCount() << std::endl;using LabelType =unsigned short;using ShapeLabelObjectType = itk::ShapeLabelObject< LabelType, Dimension >;using LabelMapType = itk::LabelMap< ShapeLabelObjectType >;//標簽using LabelShapeKeepNObjectsImageFilterType = itk::LabelShapeKeepNObjectsImageFilter<OutputImageType>;LabelShapeKeepNObjectsImageFilterType::Pointer labelShapeKeepNObjectsImageFilter =LabelShapeKeepNObjectsImageFilterType::New();labelShapeKeepNObjectsImageFilter->SetInput(connected->GetOutput());labelShapeKeepNObjectsImageFilter->SetBackgroundValue(0);labelShapeKeepNObjectsImageFilter->SetNumberOfObjects(5);//labelShapeKeepNObjectsImageFilter->SetReverseOrdering(true);labelShapeKeepNObjectsImageFilter->SetAttribute(LabelShapeKeepNObjectsImageFilterType::LabelObjectType::NUMBER_OF_PIXELS);labelShapeKeepNObjectsImageFilter->Update();//轉換maptypedef itk::LabelImageToShapeLabelMapFilter< OutputImageType, LabelMapType> I2LType;I2LType::Pointer i2l = I2LType::New();i2l->SetInput(labelShapeKeepNObjectsImageFilter->GetOutput());i2l->SetComputePerimeter(true);i2l->Update();LabelMapType* labelMap = i2l->GetOutput();std::cout << "File " << "\"" << fileName << "\""<< " has " << labelMap->GetNumberOfLabelObjects() << " labels." << std::endl;// Retrieve all attributesfor (unsigned int n = 0; n < labelMap->GetNumberOfLabelObjects(); ++n){ShapeLabelObjectType* labelObject = labelMap->GetNthLabelObject(n);std::cout << n<<" Label: "<< itk::NumericTraits<LabelMapType::LabelType>::PrintType(labelObject->GetLabel()) << std::endl; std::cout << " NumberOfPixels: "<< labelObject->GetNumberOfPixels() << std::endl;std::cout << " BoundingBox: "<< labelObject->GetBoundingBox() << std::endl;std::cout << " PhysicalSize: "<< labelObject->GetPhysicalSize() << std::endl;std::cout << " Centroid: "<< labelObject->GetCentroid() << std::endl;std::cout << " NumberOfPixelsOnBorder: "<< labelObject->GetNumberOfPixelsOnBorder() << std::endl;std::cout << " PerimeterOnBorder: "<< labelObject->GetPerimeterOnBorder() << std::endl;std::cout << " FeretDiameter: "<< labelObject->GetFeretDiameter() << std::endl;std::cout << " PrincipalMoments: "<< labelObject->GetPrincipalMoments() << std::endl;std::cout << " PrincipalAxes: "<< labelObject->GetPrincipalAxes() << std::endl;std::cout << " Elongation: "<< labelObject->GetElongation() << std::endl;std::cout << " Perimeter: "<< labelObject->GetPerimeter() << std::endl;std::cout << " Roundness: "<< labelObject->GetRoundness() << std::endl;std::cout << " EquivalentSphericalRadius: "<< labelObject->GetEquivalentSphericalRadius() << std::endl;std::cout << " EquivalentSphericalPerimeter: "<< labelObject->GetEquivalentSphericalPerimeter() << std::endl;std::cout << " EquivalentEllipsoidDiameter: "<< labelObject->GetEquivalentEllipsoidDiameter() << std::endl; std::cout << " OrientedBoundingBoxOrigin: "<< labelObject->GetOrientedBoundingBoxOrigin() << std::endl;std::cout << " Flatness: "<< labelObject->GetFlatness() << std::endl;std::cout << " PerimeterOnBorderRatio: "<< labelObject->GetPerimeterOnBorderRatio() << std::endl<<std::endl<<std::endl;std::cout<< std::endl << std::endl << std::endl;}//查找最大sizeint maxsize = 0;for (unsigned int n = 0; n < 5; ++n)//注意越界{ShapeLabelObjectType* labelObject = labelMap->GetNthLabelObject(n);if (labelObject->GetNumberOfPixels() > maxsize){maxsize = labelObject->GetNumberOfPixels();}}std::cout << "maxsize= " << maxsize << std::endl;//小于0.1倍最大size的丟棄int countToDel = 0;for (unsigned int n = 0; n < 5; ++n)//注意越界{ShapeLabelObjectType* labelObject = labelMap->GetNthLabelObject(n);if (labelObject->GetNumberOfPixels() < 0.1* maxsize){++countToDel;std::cout << "delete..."<<labelObject->GetLabel() << std::endl;}}labelShapeKeepNObjectsImageFilter->SetNumberOfObjects(5-countToDel);//著色using RescaleFilterType = itk::RescaleIntensityImageFilter<OutputImageType, ImageType>;RescaleFilterType::Pointer rescaleFilter = RescaleFilterType::New();rescaleFilter->SetOutputMinimum(0);rescaleFilter->SetOutputMaximum(255);rescaleFilter->SetInput(labelShapeKeepNObjectsImageFilter->GetOutput());#ifdef ENABLE_QUICKVIEWQuickView viewer;viewer.AddImage(reader->GetOutput(), true, itksys::SystemTools::GetFilenameName(fileName));std::stringstream desc;desc << "Largest object of " << connected->GetObjectCount() << " objects";viewer.AddImage(rescaleFilter->GetOutput(), true, desc.str());viewer.Visualize();#endifreturn EXIT_SUCCESS; }

對左邊圖像提取最大的5個連通區域，并排除size小于0.1最大size的連通區域的結果，只剩下3個連通區域。

總結

以上是生活随笔為你收集整理的二维ITK连通区域提取的全部內容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網站內容還不錯，歡迎將生活随笔推薦給好友。