private static void CollectTrainingExamples(OCR.ShapeNet ShapeNet, List <Example> Examples)
{
List <String> Filenames = new List <string>(0);
Example newExample;
int NumberExamples = ShapeNet.lNumberExamples;
//init learning data
//collect all available examples
for (int lIndex = 0; lIndex < ShapeNet.ShapeList.Count; lIndex++)
{
Filenames.Clear();
MakeExampleList(Filenames, ShapeNet.ShapeList[lIndex].SampleFolder);
if (Filenames.Count > 0)
{
for (long lIndex2 = 0; lIndex2 < NumberExamples; lIndex2++)
{
newExample = new Example();
newExample.Filename = Filenames[_r.Next(Filenames.Count)];
newExample.ShapeId = lIndex;
Examples.Add(newExample);
}
}
}
}
private void CreateNetwork(OCR.ShapeNet ShapeNet)
{
ShapeNet.NeuralNetwork.ClearNetwork();
int OutputNodes = ShapeNet.ShapeList.Count;
int Layer1Nodes = Convert.ToInt32(txtNodesLayer1.Text);
int Layer2Nodes = Convert.ToInt32(txtNodesLayer2.Text);
//fill the network with nodes
for (int lIndex = 0; lIndex < 48; lIndex++)
{
ShapeNet.NeuralNetwork.AddNode(0, eNeuralNodeType.eInput);
}
for (int lIndex = 0; lIndex < Layer1Nodes; lIndex++)
{
ShapeNet.NeuralNetwork.AddNode(1, eNeuralNodeType.eHidden);
}
for (int lIndex = 0; lIndex < Layer2Nodes; lIndex++)
{
ShapeNet.NeuralNetwork.AddNode(2, eNeuralNodeType.eHidden);
}
for (int lIndex = 0; lIndex < OutputNodes; lIndex++)
{
ShapeNet.NeuralNetwork.AddNode(3, eNeuralNodeType.eOutput);
}
ShapeNet.NeuralNetwork.InitNetworkForLearning();
}
/// <summary>
/// This function checks if this component is recognised as a connected component, meaning that
/// it may consist of multiple characters.
/// Rules
/// 1. Recognised as connected
/// 2. Low probability in recognition
/// </summary>
public void CheckAndRepairConnected(ShapeNet ShapeNet)
{
if (CouldBeConnected)
{
RecogniseComponent.NumberConnectedComponents++;
if (DebugTrace.DebugTrace.ApplySplitOnConnectedComponents)
{
Split(ShapeNet);
}
}
}
private void FillListView(OCR.ShapeNet SelectedShapeNet)
{
lvShapes.SelectedItems.Clear();
lvShapes.ItemsSource = SelectedShapeNet.ShapeList;
lvShapes.Items.Refresh();
//foreach (OCR.ShapeListEntry Shape in SelectedShapeNet.ShapeList)
//{
// lvShapes.Items.Add(Shape);
//}
}
/// <summary>
/// This functions 'recognises' the given compontent by running its features through
/// a neural network.
/// </summary>
/// <param name="Component"></param>
public static void RecogniseWithoutConnectedRepair(ShapeNet ShapeNet, PageComponent Component)
{
sNeuralOutput Result = new sNeuralOutput();
ShapeNet.NeuralNetwork.ComputeOutput(RecogniseComponent.CalculateNetworkInput(Component), Result);
for (int i = 0; i < Result.fOutputs.Count; i++)
{
if (Result.fOutputs[i] >= 0.01)
{
Component.AddRecognitionResult(Result.fOutputs[i], ShapeNet.ShapeList[i].Shape);
}
}
}
private static void CreateTrainingData(OCR.ShapeNet ShapeNet, List <Example> Examples)
{
NeuralNetwork.LearningData oLearningData;
OCR.PageComponent PageComponent;
sNeuralInput newInput;
foreach (Example Example in Examples)
{
PageComponent = new OCR.PageComponent();
PageComponent.LoadBitmap(Example.Filename);
ExtractFeatures.ExecuteExtractFeatures(PageComponent, true);
PageComponent.Position.Ascent = ShapeNet.ShapeList[Example.ShapeId].Position.Ascent;
PageComponent.Position.Height = ShapeNet.ShapeList[Example.ShapeId].Position.Height;
PageComponent.Position.Center = ShapeNet.ShapeList[Example.ShapeId].Position.Center;
PageComponent.Position.Base = ShapeNet.ShapeList[Example.ShapeId].Position.Base;
PageComponent.Position.Descent = ShapeNet.ShapeList[Example.ShapeId].Position.Descent;
//Fill the
oLearningData = new NeuralNetwork.LearningData();
oLearningData.oInput.fInputs.Clear();
oLearningData.oOutput.fOutputs.Clear();
newInput = RecogniseComponent.CalculateNetworkInput(PageComponent);
oLearningData.oInput = newInput;
for (long lIndex2 = 0; lIndex2 < ShapeNet.ShapeList.Count; lIndex2++)
{
if (Example.ShapeId == lIndex2)
{
oLearningData.oOutput.fOutputs.Add(1);
}
else
{
oLearningData.oOutput.fOutputs.Add(0);
}
}
ShapeNet.NeuralNetwork.AddSituation(oLearningData);
}
ShapeNet.NeuralNetwork.ComputeInputRatios();
}
/// <summary>
/// This functions 'recognises' the given compontent by running its features through
/// a neural network.
/// </summary>
/// <param name="Component"></param>
public static void Recognise(ShapeNet ShapeNet, PageComponent Component)
{
RecogniseWithoutConnectedRepair(ShapeNet, Component);
Component.CheckAndRepairConnected(ShapeNet);
}
private void FillShapeList(OCR.ShapeNet SelectedShapeNet)
{
lstShapes.UnselectAll();
lstShapes.ItemsSource = SelectedShapeNet.ShapeList;
lstShapes.Items.Refresh();
}
public void DoFillFileList(string Filter, OCR.ShapeNet ShapeNet)
{
try
{
if (ShapeFiles == null)
{
ShapeFiles = new List <ShapeFileList>(0);
}
if (ShapeFiles.Count == 0)
{
DirectoryInfo di = new DirectoryInfo("D:\\OCR\\Images\\");
FileInfo[] Files = di.GetFiles("*_org.bmp");
foreach (FileInfo fi in Files)
{
if (Filter.Length > 0)
{
String Filename;
RecognitionResult MostLikelyShape;
// OCR.ShapeNet Shapenet = new OCR.ShapeNet();
PageComponent Component = new PageComponent();
//Get the selected item
Filename = fi.Name.ToString();
//Load the image
Component.LoadBitmap(fi.FullName.ToString());
//Extract features and recognize
ExtractFeatures.ExecuteExtractFeatures(Component, true);
RecogniseComponent.Recognise(ShapeNet, Component);
//Fill the recognition list
MostLikelyShape = new RecognitionResult();
MostLikelyShape.Probability = 0;
foreach (RecognitionResult Result in Component.RecognitionResults)
{
if (Result.Probability > MostLikelyShape.Probability)
{
MostLikelyShape = Result;
}
}
//Add ShapeFile to cache
ShapeFileList ShapeFile;
ShapeFile.Shape = MostLikelyShape.Content;
ShapeFile.File = fi.Name.ToString();
ShapeFiles.Add(ShapeFile);
//Add file to filelist
if (MostLikelyShape.Content == Filter)
{
bwFileList.ReportProgress(0, fi.Name.ToString());
}
}
else
{
bwFileList.ReportProgress(0, fi.Name.ToString());
}
}
}
else
{
foreach (ShapeFileList ShapeFile in ShapeFiles)
{
if (Filter.Length == 0)
{
bwFileList.ReportProgress(0, ShapeFile.File);
}
else
{
if (Filter == ShapeFile.Shape)
{
bwFileList.ReportProgress(0, ShapeFile.File);
}
}
}
}
}
catch (Exception e)
{
Console.WriteLine("Exception caught: " + e.Message);
Console.WriteLine(" In: " + e.StackTrace);
}
}
/// <summary>
/// This functions splits a character in multiple components. Used for splitting connected characters
/// </summary>
public void Split(ShapeNet ShapeNet)
{
List <int> SplitLines;
//Step 1: Get the position for possible splits
SplitLines = SplitDeterminePositions();
if (SplitLines.Count == 0)
{
return;
}
RecognitionResults.Clear();
//Step 2: Find the combination of the best (highest scores) recognised components
List <PageComponent> Components = new List <PageComponent>(0);
PageComponent newComponent;
PageComponent prevComponent;
Rectangle SplitArea;
int start, end;
SplitLines.Insert(0, 0);
SplitLines.Add(Width - 1);
start = 0;
end = 1;
while (end < SplitLines.Count)
{
SplitArea = new Rectangle(SplitLines[start] + 1, 0, SplitLines[end] - SplitLines[start] - 2, Height);
if (SplitArea.Width > 0 && SplitArea.Height > 0)
{
while (NumberPixelsOnRow(BinaryBytes, SplitArea, 0, 0) == 0)
{
SplitArea.Y = SplitArea.Y + 1;
SplitArea.Height = SplitArea.Height - 1;
}
while (NumberPixelsOnRow(BinaryBytes, SplitArea, SplitArea.Height - 1, 0) == 0)
{
SplitArea.Height = SplitArea.Height - 1;
}
newComponent = PartialCopy(SplitArea);
ExtractFeatures.ExecuteExtractFeatures(newComponent, false);
RecogniseComponent.RecogniseWithoutConnectedRepair(ShapeNet, newComponent);
if (Components.Count > 0 && end - start > 1)
{
prevComponent = Components.Last();
if (prevComponent.ContentProbability < newComponent.ContentProbability && newComponent.Content != "connected" && newComponent.Content != "garbage")
{
Components.Remove(prevComponent);
Components.Add(newComponent);
}
else
{
start = end - 1;
end--;
}
}
else
{
Components.Add(newComponent);
}
}
end++;
}
//Add the new recognition result
RecognitionResult newResult;
newResult = new RecognitionResult();
newResult.Content = "";
newResult.Probability = 0;
foreach (PageComponent Component in Components)
{
newResult.Content += Component.Content;
newResult.Probability += Component.ContentProbability;
}
newResult.Probability = newResult.Probability / Components.Count;
RecognitionResults.Add(newResult);
//Save a copy of the image to the disc
if (DebugTrace.DebugTrace.TraceFeatures)
{
String ComponentID = "000000" + ID;
ComponentID = ComponentID.Substring(ComponentID.Length - 6);
foreach (int SplitLine in SplitLines)
{
int pointer = SplitLine;
for (int y = 0; y < Height; y++)
{
if (BinaryBytes[SplitLine, y] == 0xFF)
{
BinaryBytes[SplitLine, y] = 0x10;
}
pointer += Stride;
}
}
Bitmap Bitmap = DebugTrace.DebugTrace.CreateBitmapFromByteArray(BinaryBytes, new Size(Width, Height));
String Filename = DebugTrace.DebugTrace.TraceFeatureFolder + "image_" + ComponentID + "_split.bmp";
Bitmap.Save(Filename);
}
}