//JAVA TO C# CONVERTER WARNING: Method 'throws' clauses are not available in .NET:
//ORIGINAL LINE: public static void main(String[] args) throws java.io.IOException
public static void Main(string[] args)
{
// path to image directory
string imageDir = "/home/zoran/Downloads/MihailoHSLTest/trening";
// image names - used for output neuron labels
List <string> imageLabels = new ArrayList();
imageLabels.Add("bird");
imageLabels.Add("cat");
imageLabels.Add("dog");
// create dataset
IDictionary <string, FractionRgbData> map = ImageRecognitionHelper.getFractionRgbDataForDirectory(new File(imageDir), new Dimension(20, 20));
DataSet dataSet = ImageRecognitionHelper.createRGBTrainingSet(imageLabels, map);
// create neural network
List <int?> hiddenLayers = new List <int?>();
hiddenLayers.Add(12);
NeuralNetwork nnet = ImageRecognitionHelper.createNewNeuralNetwork("someNetworkName", new Dimension(20, 20), ColorMode.COLOR_RGB, imageLabels, hiddenLayers, TransferFunctionType.SIGMOID);
// set learning rule parameters
MomentumBackpropagation mb = (MomentumBackpropagation)nnet.LearningRule;
mb.LearningRate = 0.2;
mb.MaxError = 0.9;
mb.Momentum = 1;
// traiin network
Console.WriteLine("NNet start learning...");
nnet.learn(dataSet);
Console.WriteLine("NNet learned");
}
/// <summary>
/// Creates neural network for OCR, which contains OCR plugin. OCR plugin provides interface for character recognition. </summary>
/// <param name="label"> neural network label </param>
/// <param name="samplingResolution"> character size in pixels (all characters will be scaled to this dimensions during recognition) </param>
/// <param name="colorMode"> color mode used fr recognition </param>
/// <param name="characterLabels"> character labels for output neurons </param>
/// <param name="layersNeuronsCount"> number of neurons ih hidden layers </param>
/// <param name="transferFunctionType"> neurons transfer function type </param>
/// <returns> returns NeuralNetwork with the OCR plugin </returns>
public static NeuralNetwork createNewNeuralNetwork(string label, Dimension samplingResolution, ColorMode colorMode, List <string> characterLabels, List <int?> layersNeuronsCount, TransferFunctionType transferFunctionType)
{
NeuralNetwork neuralNetwork = ImageRecognitionHelper.createNewNeuralNetwork(label, samplingResolution, colorMode, characterLabels, layersNeuronsCount, transferFunctionType);
neuralNetwork.addPlugin(new OcrPlugin(samplingResolution, colorMode));
return(neuralNetwork);
}
//JAVA TO C# CONVERTER WARNING: Method 'throws' clauses are not available in .NET:
//ORIGINAL LINE: public static void main(String[] args) throws java.io.IOException
public static void Main(string[] args)
{
// User input parameteres
//*******************************************************************************************************************************
string imagePath = "C:/Users/Mihailo/Desktop/OCR/slova.png"; //path to the image with letters *
string folderPath = "C:/Users/Mihailo/Desktop/OCR/ImagesDir/"; // loaction folder for storing segmented letters *
string textPath = "C:/Users/Mihailo/Desktop/OCR/slova.txt"; // path to the .txt file with text on the image *
string networkPath = "C:/Users/Mihailo/Desktop/OCR/network.nnet"; // location where the network will be stored *
int fontSize = 12; // fontSize, predicted by height of the letters, minimum font size is 12 pt *
int scanQuality = 300; // scan quality, minimum quality is 300 dpi *
//*******************************************************************************************************************************
BufferedImage image = ImageIO.read(new File(imagePath));
ImageFilterChain chain = new ImageFilterChain();
chain.addFilter(new GrayscaleFilter());
chain.addFilter(new OtsuBinarizeFilter());
BufferedImage binarizedImage = chain.processImage(image);
Letter letterInfo = new Letter(scanQuality, binarizedImage);
// letterInfo.recognizeDots(); // call this method only if you want to recognize dots and other litle characters, TODO
Text texTInfo = new Text(binarizedImage, letterInfo);
OCRTraining ocrTraining = new OCRTraining(letterInfo, texTInfo);
ocrTraining.FolderPath = folderPath;
ocrTraining.TrainingTextPath = textPath;
ocrTraining.prepareTrainingSet();
List <string> characterLabels = ocrTraining.CharacterLabels;
IDictionary <string, FractionRgbData> map = ImageRecognitionHelper.getFractionRgbDataForDirectory(new File(folderPath), new Dimension(20, 20));
DataSet dataSet = ImageRecognitionHelper.createBlackAndWhiteTrainingSet(characterLabels, map);
dataSet.FilePath = "C:/Users/Mihailo/Desktop/OCR/DataSet1.tset";
dataSet.save();
List <int?> hiddenLayers = new List <int?>();
hiddenLayers.Add(12);
NeuralNetwork nnet = ImageRecognitionHelper.createNewNeuralNetwork("someNetworkName", new Dimension(20, 20), ColorMode.BLACK_AND_WHITE, characterLabels, hiddenLayers, TransferFunctionType.SIGMOID);
BackPropagation bp = (BackPropagation)nnet.LearningRule;
bp.LearningRate = 0.3;
bp.MaxError = 0.1;
// MultiLayerPerceptron mlp = new MultiLayerPerceptron(12,13);
// mlp.setOutputNeurons(null);
Console.WriteLine("Start learning...");
nnet.learn(dataSet);
Console.WriteLine("NNet learned");
nnet.save(networkPath);
}
/// <summary>
/// Create training set
/// </summary>
/// <param name="imageWithChars"> </param>
/// <param name="chars"> </param>
/// <param name="scaleToDim"> </param>
/// <param name="trainingSetName"> </param>
public static DataSet createTrainingSet(string trainingSetName, BufferedImage imageWithChars, string chars, Dimension scaleToDim, List <string> imageLabels)
{
// convert chars from string to list
List <string> charList = Arrays.asList(chars.Split(" ", true)); // izgleda da ovo zeza...
// extract individual char images which will be used to create training set
CharExtractor charExtractor = new CharExtractor(imageWithChars);
Dictionary <string, BufferedImage> charImageMap = charExtractor.extractCharImagesToLearn(imageWithChars, charList, scaleToDim);
// prepare image labels (we need them to label output neurons...)
// ArrayList<String> imageLabels = new ArrayList<String>();
foreach (string imgName in charImageMap.Keys)
{
StringTokenizer st = new StringTokenizer(imgName, "._");
string imgLabel = st.nextToken();
if (!imageLabels.Contains(imgLabel)) // check for duplicates ...
{
imageLabels.Add(imgLabel);
}
}
imageLabels.Sort();
// get RGB image data - map chars and their their rgb data
IDictionary <string, FractionRgbData> imageRgbData = ImageUtilities.getFractionRgbDataForImages(charImageMap);
// also put junk all black and white image in training set (for black n whit emode)
BufferedImage allWhite = new BufferedImage(scaleToDim.Width, scaleToDim.Height, BufferedImage.TYPE_INT_RGB);
Graphics g = allWhite.Graphics;
g.Color = Color.WHITE;
g.fillRect(0, 0, allWhite.Width, allWhite.Height);
imageRgbData["allWhite"] = new FractionRgbData(allWhite);
// BufferedImage allBlack = new BufferedImage(charDimension.getWidth(), charDimension.getHeight(), BufferedImage.TYPE_INT_RGB);
// g = allBlack.getGraphics();
// g.setColor(Color.BLACK);
// g.fillRect(0, 0, allBlack.getWidth(), allBlack.getHeight());
// imageRgbData.put("allBlack", new FractionRgbData(allBlack));
// put junk images (all red, blue, green) to avoid errors (used for full color mode)
// BufferedImage allRed = new BufferedImage(charDimension.getWidth(), charDimension.getHeight(), BufferedImage.TYPE_INT_RGB);
// Graphics g = allRed.getGraphics();
// g.setColor(Color.RED);
// g.fillRect(0, 0, allRed.getWidth(), allRed.getHeight());
// imageRgbData.put("allRed", new FractionRgbData(allRed));
//
// BufferedImage allBlue = new BufferedImage(charDimension.getWidth(), charDimension.getHeight(), BufferedImage.TYPE_INT_RGB);
// g = allBlue.getGraphics();
// g.setColor(Color.BLUE);
// g.fillRect(0, 0, allBlue.getWidth(), allBlue.getHeight());
// imageRgbData.put("allBlue", new FractionRgbData(allBlue));
//
// BufferedImage allGreen = new BufferedImage(charDimension.getWidth(), charDimension.getHeight(), BufferedImage.TYPE_INT_RGB);
// g = allGreen.getGraphics();
// g.setColor(Color.GREEN);
// g.fillRect(0, 0, allGreen.getWidth(), allGreen.getHeight());
// imageRgbData.put("allGreen", new FractionRgbData(allGreen));
// create training set using image rgb data
DataSet dataSet = ImageRecognitionHelper.createBlackAndWhiteTrainingSet(imageLabels, imageRgbData);
//DataSet dataSet = ImageRecognitionHelper.createTrainingSet(this.imageLabels, imageRgbData);
dataSet.Label = trainingSetName;
return(dataSet);
}
