internal OcrReader(NeuralNet descriptions, NeuralNet tables, NeuralNet numbers, NeuralNet headlines, NeuralNet delimiters)
{
nnDescriptions = descriptions;
nnTables = tables;
nnNumbers = numbers;
nnHeadlines = headlines;
nnDelimiters = delimiters;
tableConfig = TableItem.Load(PathHelpers.BuildConfigFilename("TableItems"));
descriptionConfig = DescriptionItem.Load(PathHelpers.BuildConfigFilename("Descriptions"));
itemConfig = ItemValues.Load(PathHelpers.BuildConfigFilename("ItemValues"));
wordList = File.ReadAllLines(PathHelpers.BuildWordFilename("words"));
Dictionary<string, int> lookup = new Dictionary<string, int>();
foreach (DescriptionItem di in descriptionConfig)
{
string[] split = di.Description.Split(new char[] { ' ' });
foreach (string s in split)
{
lookup[s] = 1;
}
}
List<string> tmp = new List<string>(lookup.Keys);
wordListDescription = tmp.ToArray();
}
// Train the neural networks.
//Note: This depends on the way your knowlwdge directory is set up and may need to be changed whenever retraining!
public static void TrainNN(out OcrReader ocrReader)
{
int dimensionX = Properties.Settings.Default.DimensionX;
int dimensionY = Properties.Settings.Default.DimensionY;
NeuralNet nnDescriptions, nnTables, nnNumbers, nnHeadlines, nnDelimiters;
List<string> knowledge = new List<string>();
List<char> netKeys = new List<char>();
string files = "abcdefghijklmnopqrstuvwy";
for (int i = 0; i < files.Length; i++)
{
knowledge.Add(PathHelpers.BuildKnowledgeFilename(DescriptionsNetwork, files[i].ToString() + "_lower"));
netKeys.Add(files[i]);
}
knowledge.Add(PathHelpers.BuildKnowledgeFilename(DescriptionsNetwork, "delimiter"));
netKeys.Add('.');
netKeys.Add('#');
nnDescriptions = new NeuralNet(dimensionX, dimensionY, netKeys, knowledge);
nnDescriptions.SaveFile = PathHelpers.BuildNetworkFilename(DescriptionsNetwork);
nnDescriptions.Train(Properties.Settings.Default.SamplesDescriptions);
knowledge.Clear();
netKeys.Clear();
files = "ABCDEFGHIJKLMNOPRSTUVWXY";
for (int i = 0; i < files.Length; i++)
{
knowledge.Add(PathHelpers.BuildKnowledgeFilename(TablesNetwork, files[i].ToString() + "_upper"));
netKeys.Add(files[i]);
}
knowledge.Add(PathHelpers.BuildKnowledgeFilename(TablesNetwork, "delimiter"));
netKeys.Add(':');
netKeys.Add('(');
netKeys.Add(')');
knowledge.Add(PathHelpers.BuildKnowledgeFilename(TablesNetwork, "minus"));
netKeys.Add('-');
nnTables = new NeuralNet(dimensionX, dimensionY, netKeys, knowledge);
nnTables.SaveFile = PathHelpers.BuildNetworkFilename(TablesNetwork);
nnTables.Train(Properties.Settings.Default.SamplesTables);
knowledge.Clear();
netKeys.Clear();
knowledge.Add(PathHelpers.BuildKnowledgeFilename(DelimitersNetwork, "comma"));
netKeys.Add('#');
knowledge.Add(PathHelpers.BuildKnowledgeFilename(DelimitersNetwork, "dot"));
netKeys.Add('.');
knowledge.Add(PathHelpers.BuildKnowledgeFilename(DelimitersNetwork, "minus"));
netKeys.Add('-');
nnDelimiters = new NeuralNet(dimensionX, dimensionY, netKeys, knowledge);
nnDelimiters.SaveFile = PathHelpers.BuildNetworkFilename(DelimitersNetwork);
nnDelimiters.Train(Properties.Settings.Default.SamplesNumbers);
knowledge.Clear();
netKeys.Clear();
files = "0123456789";
for (int i = 0; i < files.Length; i++)
{
knowledge.Add(PathHelpers.BuildKnowledgeFilename(NumbersNetwork, files[i].ToString()));
netKeys.Add(files[i]);
}
nnNumbers = new NeuralNet(dimensionX, dimensionY, netKeys, knowledge);
nnNumbers.SaveFile = PathHelpers.BuildNetworkFilename(NumbersNetwork);
nnNumbers.Train(Properties.Settings.Default.SamplesNumbers);
knowledge.Clear();
netKeys.Clear();
files = "0123456789";
for (int i = 0; i < files.Length; i++)
{
knowledge.Add(PathHelpers.BuildKnowledgeFilename(HeadlinesNetwork, files[i].ToString()));
netKeys.Add(files[i]);
}
files = "ABCDEFGHIJKLMNOPQRSTUVXYZ";
for (int i = 0; i < files.Length; i++)
{
knowledge.Add(PathHelpers.BuildKnowledgeFilename(HeadlinesNetwork, files[i].ToString() + "_upper"));
netKeys.Add(files[i]);
}
files = "abcdefghijklmnopqrstuvwyz";
for (int i = 0; i < files.Length; i++)
{
knowledge.Add(PathHelpers.BuildKnowledgeFilename(HeadlinesNetwork, files[i].ToString() + "_lower"));
netKeys.Add(files[i]);
}
knowledge.Add(PathHelpers.BuildKnowledgeFilename(HeadlinesNetwork, "minus"));
netKeys.Add('-');
nnHeadlines = new NeuralNet(15, 22, netKeys, knowledge);
nnHeadlines.SaveFile = PathHelpers.BuildNetworkFilename(HeadlinesNetwork);
nnHeadlines.Factor = 2;
nnHeadlines.Train(Properties.Settings.Default.SamplesHeadlines);
ocrReader = new OcrReader(nnDescriptions, nnTables, nnNumbers,nnHeadlines, nnDelimiters);
}
