public void CheckOne()
{
ANeuralNetwork network = new ANeuralNetwork();
network.Load("..\\..\\..\\..\\savedData.txt");
var sampleDirs = Directory.GetDirectories("..\\..\\..\\..\\Data\\TestData\\").ToList();
int fCount = Directory.GetFiles("..\\..\\..\\..\\Data\\GroundData\\", "*.png", SearchOption.AllDirectories).Length;
var poliManager = new PolynomialManager();
poliManager.InitBasis(15, 100);
double count = 0;
foreach (var dir in sampleDirs)
{
var files = Directory.GetFiles(dir, "*.png", SearchOption.TopDirectoryOnly).ToList();
var index = new Random().Next(files.Count);
var file = files[index];
var value = Convert.ToInt32(new StreamReader(dir + "\\" + "value.txt").ReadToEnd());
++count;
ComplexMoments tmpMoments;
ProcessOneImage(file, poliManager, 100, out tmpMoments);
CvInvoke.Imshow(dir, tmpMoments.Real);
var tmpInput = tmpMoments.ToListOfDouble();
var output = network.Predict(tmpInput);
var predictedValue = Convert.ToInt32(output.IndexOf(output.Max()));
Console.WriteLine(predictedValue);
CvInvoke.WaitKey();
}
}
static void Main(string[] args)
{
string key;
do
{
Console.WriteLine("===Enter next values to do something:===");
Console.WriteLine(" '1' - to generate data.");
Console.WriteLine(" '2' - to train network.");
Console.WriteLine(" '3' - to check recognizing precision.");
Console.WriteLine(" '4' - to recognize single image.");
Console.WriteLine(" 'exit' - to close the application.");
Console.WriteLine();
key = Console.ReadLine();
var pm = new ProcessMoments();
if (key == "1")
{
var poliManager = new PolynomialManager();
var pd = new ProcessData();
poliManager.InitBasis(15, 100);
pd.DistributeData("..\\..\\..\\..\\Data\\LabeledData\\", "..\\..\\..\\..\\Data\\GroundData\\", "..\\..\\..\\..\\Data\\TestData\\", 50);
var dictionary = pm.GenerateMoments("..\\..\\..\\..\\Data\\GroundData\\", 100, poliManager);
pd.SaveMoments("..\\..\\..\\..\\Moments.yaml", dictionary);
}
else if (key == "2")
{
var pd = new ProcessData();
var tmpDict = new SortedDictionary <string, List <ComplexMoments> >();
pd.ReadMoments("..\\..\\..\\..\\Moments.yaml", tmpDict);
var layers = new List <uint> {
225, 160, 60, 25, 9
};
pm.Train(layers, tmpDict, 100000, 0.4, 0.01);
}
else if (key == "3")
{
pm.Check();
}
else if (key == "4")
{
pm.CheckOne();
}
} while (key != "exit");
}
public void ProcessOneImage(string imagePath, PolynomialManager polyManager, int diameter, out ComplexMoments res)
{
var image = CvInvoke.Imread(imagePath, ImreadModes.Grayscale);
if (image.IsEmpty)
{
throw new Exception("Empty image");
}
CvInvoke.Threshold(image, image, 127, 255, ThresholdType.BinaryInv);
var blobs = polyManager.DetectBlobs(image);
if (blobs.Count != 1)
{
throw new Exception("Incorrect input data. More then one blob.");
}
var nblobs = polyManager.NormalizeBlobs(blobs, diameter);
res = polyManager.Decompose(nblobs[0]);
}
public void Check()
{
ANeuralNetwork network = new ANeuralNetwork();
network.Load("..\\..\\..\\..\\savedData.txt");
var sampleDirs = Directory.GetDirectories("..\\..\\..\\..\\Data\\TestData\\").ToList();
int fCount = Directory.GetFiles("..\\..\\..\\..\\Data\\GroundData\\", "*.png", SearchOption.AllDirectories).Length;
double precision = 0;
var poliManager = new PolynomialManager();
poliManager.InitBasis(15, 100);
double count = 0;
double trueCount = 0;
foreach (var dir in sampleDirs)
{
var files = Directory.GetFiles(dir, "*.png", SearchOption.TopDirectoryOnly).ToList();
var value = Convert.ToInt32(new StreamReader(dir + "\\" + "value.txt").ReadToEnd());
foreach (var file in files)
{
++count;
ComplexMoments tmpMoments;
ProcessOneImage(file, poliManager, 100, out tmpMoments);
var tmpInput = tmpMoments.ToListOfDouble();
var output = network.Predict(tmpInput);
var predictedValue = Convert.ToInt32(output.IndexOf(output.Max()));
if (predictedValue == value)
{
//++trueCount;
precision += Convert.ToDouble(100 / (double)fCount);
}
}
Console.WriteLine("Точность " + precision + "%");
//Console.WriteLine("Точность " + trueCount/count + "%");
}
}
public static void Main()
{
Console.WriteLine("Hello world");
Console.WriteLine("Max polinom:");
var max = Convert.ToInt32(Console.ReadLine());
Console.WriteLine();
Console.WriteLine("Diametr:");
var diameter = Convert.ToInt32(Console.ReadLine());
Console.WriteLine();
var pm = new PolynomialManager();
pm.InitBasis(max, diameter);
Visualisation.ShowPolynomials("Basic:", Polynomials);
var image = CvInvoke.Imread("..\\..\\..\\..\\Picture.png", ImreadModes.Grayscale);
CvInvoke.Imshow("Picture:", image);
CvInvoke.WaitKey();
var blobs = pm.DetectBlobs(image);
blobs = pm.NormalizeBlobs(blobs, diameter);
var decomposedBlobs = new List <ComplexMoments>(blobs.Count);
var recoveredBlobs = new List <Mat>(blobs.Count);
for (var i = 0; i < blobs.Count; i++)
{
decomposedBlobs.Add(pm.Decompose(blobs[i]));
recoveredBlobs.Add(pm.Recovery(decomposedBlobs[i]));
Visualisation.ShowBlobDecomposition("Восстановленные цифры:", blobs[i], recoveredBlobs[i]);
CvInvoke.WaitKey();
}
}
public SortedDictionary <string, List <ComplexMoments> > GenerateMoments(string path, int diameter, PolynomialManager polyManager)
{
var sampleDirs = Directory.GetDirectories(path).ToList();
SortedDictionary <string, List <ComplexMoments> > result = new SortedDictionary <string, List <ComplexMoments> >();
//Перебираем все папки.
for (var i = 0; i < sampleDirs.Count; i++)
{
// Ищем файл со значением буквы.
var tmpFile = new StreamReader(sampleDirs[i] + "\\" + "value.txt");
var key = tmpFile.ReadToEnd();
tmpFile.Close();
var files = Directory.GetFiles(sampleDirs[i] + "\\", "*.png");
var tmp = new List <ComplexMoments>();
foreach (var file in files)
{
//Считываем картинку.
//Обрабатываем.
ProcessOneImage(file, polyManager, diameter, out var moment);
tmp.Add(moment);
}
//Сохраняем
result.Add(key, tmp);
}
return(result);
}
