public void TestBOWKmeansTrainer2()
{
Image <Gray, byte> box = EmguAssert.LoadImage <Gray, byte>("box.png");
Brisk detector = new Brisk(30, 3, 1.0f);
VectorOfKeyPoint kpts = new VectorOfKeyPoint();
Mat descriptors = new Mat();
detector.DetectAndCompute(box, null, kpts, descriptors, false);
Mat descriptorsF = new Mat();
descriptors.ConvertTo(descriptorsF, CvEnum.DepthType.Cv32F);
//Matrix<float> descriptorsF = descriptors.Convert<float>();
BOWKMeansTrainer trainer = new BOWKMeansTrainer(100, new MCvTermCriteria(), 3, CvEnum.KMeansInitType.PPCenters);
trainer.Add(descriptorsF);
Mat vocabulary = new Mat();
trainer.Cluster(vocabulary);
BFMatcher matcher = new BFMatcher(DistanceType.L2);
BOWImgDescriptorExtractor extractor = new BOWImgDescriptorExtractor(detector, matcher);
Mat vocabularyByte = new Mat();
vocabulary.ConvertTo(vocabularyByte, CvEnum.DepthType.Cv8U);
extractor.SetVocabulary(vocabularyByte);
Mat descriptors2 = new Mat();
extractor.Compute(box, kpts, descriptors2);
}
public void TestBOWKmeansTrainer()
{
Image <Gray, byte> box = EmguAssert.LoadImage <Gray, byte>("box.png");
SURF detector = new SURF(500);
VectorOfKeyPoint kpts = new VectorOfKeyPoint();
Mat descriptors = new Mat();
detector.DetectAndCompute(box, null, kpts, descriptors, false);
BOWKMeansTrainer trainer = new BOWKMeansTrainer(100, new MCvTermCriteria(), 3, CvEnum.KMeansInitType.PPCenters);
trainer.Add(descriptors);
Mat vocabulary = new Mat();
trainer.Cluster(vocabulary);
BFMatcher matcher = new BFMatcher(DistanceType.L2);
BOWImgDescriptorExtractor extractor = new BOWImgDescriptorExtractor(detector, matcher);
extractor.SetVocabulary(vocabulary);
Mat descriptors2 = new Mat();
extractor.Compute(box, kpts, descriptors2);
}
public void RunTest()
{
using (var descriptorExtractor = SIFT.Create(500))
//using (var descriptorMatcher = new FlannBasedMatcher(ip, sp))
using (var descriptorMatcher = new BFMatcher())
using (var img = Image("lenna.png"))
{
KeyPoint[] keypoints;
Mat dictionary;
var tc = new TermCriteria(CriteriaType.MaxIter, 100, 0.001d);
using (var bowTrainer = new BOWKMeansTrainer(200, tc, 1, KMeansFlags.PpCenters))
{
var descriptors = new Mat();
descriptorExtractor.DetectAndCompute(img, null, out keypoints, descriptors);
Mat featuresUnclustered = new Mat();
featuresUnclustered.PushBack(descriptors);
featuresUnclustered.ConvertTo(featuresUnclustered, MatType.CV_32F);
dictionary = bowTrainer.Cluster(featuresUnclustered);
}
using (var bowDE = new BOWImgDescriptorExtractor(descriptorExtractor, descriptorMatcher))
{
bowDE.SetVocabulary(dictionary);
try
{
using (Mat descriptors = new Mat())
{
descriptorExtractor.Compute(img, ref keypoints, descriptors);
descriptors.ConvertTo(descriptors, MatType.CV_32F);
bowDE.Compute(img, ref keypoints, descriptors, out var arr);
Console.WriteLine(arr.Length);
Console.WriteLine(arr[0].Length);
}
}
catch (OpenCVException ex)
{
Console.WriteLine(ex.FileName);
Console.WriteLine(ex.FuncName);
Console.WriteLine(ex.Line);
throw;
}
}
dictionary.Dispose();
}
}
public void RunTest()
{
using var descriptorExtractor = SIFT.Create(500);
using var descriptorMatcher = new BFMatcher();
using var img = Image("lenna.png");
KeyPoint[] keypoints;
Mat dictionary;
var tc = new TermCriteria(CriteriaTypes.MaxIter, 100, 0.001d);
using (var bowTrainer = new BOWKMeansTrainer(200, tc, 1, KMeansFlags.PpCenters))
{
var descriptors = new Mat();
descriptorExtractor.DetectAndCompute(img, null, out keypoints, descriptors);
using var featuresUnclustered = new Mat();
featuresUnclustered.PushBack(descriptors);
featuresUnclustered.ConvertTo(featuresUnclustered, MatType.CV_32F);
dictionary = bowTrainer.Cluster(featuresUnclustered);
}
using (var bowDe = new BOWImgDescriptorExtractor(descriptorExtractor, descriptorMatcher))
{
bowDe.SetVocabulary(dictionary);
try
{
using var descriptors = new Mat();
descriptorExtractor.Compute(img, ref keypoints, descriptors);
descriptors.ConvertTo(descriptors, MatType.CV_32F);
bowDe.Compute(img, ref keypoints, descriptors, out var arr);
testOutputHelper.WriteLine(arr.Length.ToString(CultureInfo.InvariantCulture));
testOutputHelper.WriteLine(arr[0].Length.ToString(CultureInfo.InvariantCulture));
}
catch (OpenCVException ex)
{
testOutputHelper.WriteLine(ex.FileName);
testOutputHelper.WriteLine(ex.FuncName);
testOutputHelper.WriteLine(ex.Line.ToString(CultureInfo.InvariantCulture));
throw;
}
}
dictionary.Dispose();
}
static void train_svm()
{
int n_samples = 0;
SURF surf = new SURF(400);
List <Bitmap> samples = new List <Bitmap>();
List <Tuple <Bitmap, int> > data = new List <Tuple <Bitmap, int> >();
/*
* foreach (string s in System.IO.Directory.GetFiles("mail_samples"))
* {
* Bitmap f1 = new Bitmap(s);//ImageDecoder.DecodeFromFile(s);
* data.Add(new Tuple<Bitmap, int>(f1, +1));
* }
* foreach (string s in System.IO.Directory.GetFiles("phone_icons"))
* {
* Bitmap f1 = new Bitmap(s); //ImageDecoder.DecodeFromFile(s);
* data.Add(new Tuple<Bitmap, int>(f1, -1));
* }
*/
foreach (string s in System.IO.Directory.GetFiles(@"C:\test\iphone_icon"))
{
Bitmap f1 = new Bitmap(s);//ImageDecoder.DecodeFromFile(s);
if (string.Compare(System.IO.Path.GetFileNameWithoutExtension(s), "temp_1") == 0 ||
string.Compare(System.IO.Path.GetFileNameWithoutExtension(s), "scoll_selected_icon") == 0
)
{
data.Add(new Tuple <Bitmap, int>(f1, +1));
}
else
{
data.Add(new Tuple <Bitmap, int>(f1, 0));
}
}
n_samples = data.Count;
// computr bow
Mat m = new Mat();
foreach (Tuple <Bitmap, int> v in data)
{
Image <Bgr, Byte> i = new Image <Bgr, byte>(v.Item1);
Mat ii = new Mat();
CvInvoke.CvtColor(i, ii, ColorConversion.Bgr2Gray);
MKeyPoint[] kp = surf.Detect(ii);
Mat desc = new Mat();
surf.Compute(ii, new VectorOfKeyPoint(kp), desc);
m.PushBack(desc);
}
// Create the vocabulary with KMeans.
MCvTermCriteria tc = new MCvTermCriteria(100, 0.00001);
BOWKMeansTrainer bowTrainer = new BOWKMeansTrainer(16, tc, 3, KMeansInitType.PPCenters);
bowTrainer.Add(m);
Mat voca = new Mat();
bowTrainer.Cluster(voca);
//
BFMatcher matcher = new BFMatcher(DistanceType.L2);
BOWImgDescriptorExtractor bowDex = new BOWImgDescriptorExtractor(surf, matcher);
bowDex.SetVocabulary(voca);
//
Mat tDesc = new Mat();
//Matrix<int> tLabel = new Matrix<int>(1, n_samples);
Matrix <int> tLabel = new Matrix <int>(n_samples, 1);
//foreach (Tuple<Bitmap, int> v in data)
for (int j = 0; j < data.Count; j++)
{
Image <Bgr, Byte> i = new Image <Bgr, byte>(data[j].Item1);
Mat ii = new Mat();
CvInvoke.CvtColor(i, ii, ColorConversion.Bgr2Gray);
MKeyPoint[] kp = surf.Detect(ii);
Mat desc = new Mat();
bowDex.Compute(ii, new VectorOfKeyPoint(kp), desc);
tDesc.PushBack(desc);
//tLabel[0, j] = data[j].Item2;
tLabel[j, 0] = data[j].Item2;
}
//
//SVM model = new SVM();
//model.SetKernel(Emgu.CV.ML.SVM.SvmKernelType.Linear);
//model.Type = SVM.SvmType.CSvc;
//model.C = 1;
//model.TermCriteria = new MCvTermCriteria(100, 0.00001);
SVM svm = new SVM();
svm.C = 312.5;
svm.Gamma = 0.50625000000000009;
svm.SetKernel(SVM.SvmKernelType.Rbf);
svm.Type = SVM.SvmType.CSvc;
svm.Nu = 0.5;
TrainData td = new TrainData(tDesc, Emgu.CV.ML.MlEnum.DataLayoutType.RowSample, tLabel);
bool tained = svm.TrainAuto(td);
using (FileStorage fs = new FileStorage("voca.yaml", FileStorage.Mode.Write))
{
svm.Write(fs);
fs.Write(voca, "voca");
}
//using (FileStorage fs = new FileStorage("svm.yaml", FileStorage.Mode.Write))
//{
// svm.Write(fs);
//}
//svm.Save("svm.yaml");
// test
{
//Image<Bgr, Byte> test_img = new Image<Bgr, byte>(@"C:\test\scroll_left.jpg");
Image <Bgr, Byte> test_img = new Image <Bgr, byte>(@"C:\test\iphone_icon\temp_1.jpg");
//Image<Bgr, Byte> test_img = new Image<Bgr, byte>(@"C:\projects\local\testMQ\testMQ\bin\Debug\phone_icons\icon_2.jpg");
//Image<Bgr, Byte> test_img = new Image<Bgr, byte>(@"C:\test\35928233-email-icon-on-blue-background-clean-vector.jpg");
Mat ii = new Mat();
CvInvoke.CvtColor(test_img, ii, ColorConversion.Bgr2Gray);
MKeyPoint[] kp = surf.Detect(ii);
Mat desc = new Mat();
bowDex.Compute(ii, new VectorOfKeyPoint(kp), desc);
float r = svm.Predict(desc);
}
}
/// <summary>
/// Utilizado para gerar os arquivos de treinados para o SVM
/// </summary>
public void RunTrain()
{
Image <Bgr, Byte> image;
Console.WriteLine("Quantização vetorial...");
//Chama o método responsavel por carregar as imagens de treinamento e extrair os descritores
LoadTrain();
//Obter a quantidade de caracteristicas (descritores) adicionados no BOW
int count = bowTrainer.DescriptorCount;
Console.WriteLine($"Clustering {count} características");
Mat dictionary = new Mat();
//Gera os vocabularios a partir dos descritores
bowTrainer.Cluster(dictionary);
//Cria um arquivo chamado train_clustered.yml com os vocabularios
FileStorage fs = new FileStorage($@"{Path}\SVM Datasets\train_clustered.yml", FileStorage.Mode.Write);
//Adiciona ao arquivo train_clustered.yml um label vocabulary
fs.Write(dictionary, "vocabulary");
fs.ReleaseAndGetString();
//Realiza a extração dos histogramas em forma de BOW para cada imagem do treinamento
bowDE.SetVocabulary(dictionary);
Console.WriteLine("Extraindo histogramas na forma de BOW para cada imagem");
//Labels é o nome da imagem que foi treinada, é utilizado para descobrir qual é o valor da imagem
Matrix <int> labels;
List <int> listLabels = new List <int>();
Mat trainingData = new Mat(0, dictionarySize, DepthType.Cv32F, 1);
MKeyPoint[] keypoints;
Mat bowDescriptor = new Mat();
//var notas = new List<int> { 2, 5, 10, 20, 50, 100 };
for (int i = 0; i < banknotes.Count; i++)
{
var banknote = banknotes[i];
for (int j = 1; j <= banknoteSizes[banknote]; j++)
{
var fileName = $@"{Path}\Train\{banknote} ({j}).jpg";
image = new Image <Bgr, Byte>(fileName);
keypoints = extractor.Detect(image);
bowDE.Compute(image, new VectorOfKeyPoint(keypoints), bowDescriptor);
trainingData.PushBack(bowDescriptor);
listLabels.Add(banknote);
}
}
//Cria um arquivo com os dados treinados chamado trained_data.yml
fs = new FileStorage($@"{Path}\SVM Datasets\trained_data.yml", FileStorage.Mode.Write);
//Adiciona um label data ao arquivo trained_data.yml
fs.Write(trainingData, "data");
fs.ReleaseAndGetString();
labels = new Matrix <int>(listLabels.ToArray());
//Cria o arquivo trained_labels.yml que aramaze os labels das imagens que foram treinadas
fs = new FileStorage($@"{Path}\SVM Datasets\trained_labels.yml", FileStorage.Mode.Write);
//Adiciona o label no arquivo trained_labels.yml chamado labels
fs.Write(labels.Mat, "labels");
fs.ReleaseAndGetString();
Console.WriteLine("Treinamento Concluido");
//Obs.: trained_data.yml, trained_labels.yml e train_clustered.yml são os arquivos mais importantes,
//os arquivos contem todo o conhecimento necessario para o SVM
}
