// USE CASE: Creating a Document Definition by training on a set of images
public static void Creating_a_Document_Definition_by_training_on_a_set_of_images(IEngine engine)
{
string rootFolder = SamplesFolder + "\\SampleImages\\Training\\ISBN";
string batchFolder = rootFolder + "\\_TrainingBatch";
if (System.IO.Directory.Exists(batchFolder))
{
// Delete the existing training batch
System.IO.Directory.Delete(batchFolder, true);
}
trace("Create training batch and populate it with images...");
ITrainingBatch trainingBatch = engine.CreateTrainingBatch(batchFolder, "English");
try {
ITrainingDefinition newDefinition = trainingBatch.Definitions.AddNew("ISBN");
trainingBatch.AddImageFile(rootFolder + "\\00.jpg");
trainingBatch.AddImageFile(rootFolder + "\\01.jpg");
trace("Use the first page to define document structure...");
ITrainingPage firstPage = trainingBatch.Pages[0];
// Each page must be prepared before trying to work with its layout. At this stage the page is analyzed
// and primitive image objects are extracted (which can be used as helpers in user imterface).
// Than an attempt is made to predict the page layout based on the layout of verified pages if any.
firstPage.PrepareLayout();
// At this point the user must draw boxes for fields and references. In this sample we try to emulate this
// behavior by 'looking' for known text strings and 'drawing' fields around them.
for (int j = 0; j < firstPage.ImageObjects.Count; j++)
{
ITrainingImageObject obj = firstPage.ImageObjects[j];
string text = obj.RecognizedText;
if (text == "978-1-4095-3439-6")
{
// We want to extact this field. Create a data field and define its geometry on the current page.
ITrainingField isbnField = newDefinition.Fields.AddNew("ISBN", TrainingFieldTypeEnum.TFT_Field);
firstPage.SetFieldBlock(isbnField, obj.Region);
break;
}
}
for (int j = 0; j < firstPage.ImageObjects.Count; j++)
{
ITrainingImageObject obj = firstPage.ImageObjects[j];
string text = obj.RecognizedText;
if (text == "ISBN")
{
// We want to use this text for reference. Create a reference element and define its geometry on the current page.
ITrainingField isbnTag = newDefinition.Fields.AddNew("ISBNTag", TrainingFieldTypeEnum.TFT_ReferenceText);
firstPage.SetFieldBlock(isbnTag, obj.Region);
break;
}
}
assert(newDefinition.Fields.Count == 2);
// Now that we are done with this page, mark it as verified and ready for training.
trainingBatch.SubmitPageForTraining(firstPage);
traceBegin("Verify the computed layout on the remaining pages...");
for (int i = 1; i < trainingBatch.Pages.Count; i++)
{
traceBegin(i.ToString() + "...");
ITrainingPage page = trainingBatch.Pages[i];
page.PrepareLayout();
// At this point the user must verify and correct the computed layout. In this sample we assume that
// the computed layout is correct, so we just mark the page as verified and ready for training.
trainingBatch.SubmitPageForTraining(page);
traceEnd("OK");
}
traceEnd("OK");
trace("Export to AFL...");
newDefinition.ExportToAFL(batchFolder + "\\NewTemplate.afl");
} finally {
trainingBatch.Close();
}
trace("Create document definition.");
IDocumentDefinition newDocumentDefinition = engine.CreateDocumentDefinitionFromAFL(batchFolder + "\\NewTemplate.afl", "English");
trace("Use the new document definition with FlexiCapture Processor.");
CheckTrainedDocumentDefinition(engine, newDocumentDefinition, rootFolder);
}
// USE CASE: Creating a Document Definition by training on a set of images
public static void Creating_a_Document_Definition_by_training_on_a_set_of_images_LPAREN2RPAREN(IEngine engine)
{
string rootFolder = SamplesFolder + "\\SampleImages\\Training\\ISBN";
string batchFolder = rootFolder + "\\_TrainingBatch";
if (System.IO.Directory.Exists(batchFolder))
{
// Delete the existing training batch
System.IO.Directory.Delete(batchFolder, true);
}
trace("Create training batch and populate it with images...");
ITrainingBatch trainingBatch = engine.CreateTrainingBatch(batchFolder, "English");
try {
ITrainingDefinition newDefinition = trainingBatch.Definitions.AddNew("ISBN");
trainingBatch.AddImageFile(rootFolder + "\\00.jpg");
trainingBatch.AddImageFile(rootFolder + "\\01.jpg");
traceBegin("The user iterates through the added images until all the pages have been submitted for training...");
ITrainingPage page = trainingBatch.PrepareNextPageNotSubmittedForTraining();
while (page != null)
{
traceBegin(page.ID.ToString() + "...");
// The user can 'draw' fields and references on any page while inside this loop. Any modification in the document definition will reset
// the 'verified' flag for all pages and the loop will automatically reiterate through all pages. In this sample we try to emulate the
// user 'drawing' fields on the first page.
if (page == trainingBatch.Pages[0])
{
// On the fist page we 'look' for known text strings and 'draw' fields around them
for (int j = 0; j < page.ImageObjects.Count; j++)
{
ITrainingImageObject obj = page.ImageObjects[j];
string text = obj.RecognizedText;
if (text == "978-1-4095-3439-6")
{
// We want to extact this field. Create a data field and define its geometry on the current page.
ITrainingField isbnField = newDefinition.Fields.AddNew("ISBN", TrainingFieldTypeEnum.TFT_Field);
page.SetFieldBlock(isbnField, obj.Region);
break;
}
}
for (int j = 0; j < page.ImageObjects.Count; j++)
{
ITrainingImageObject obj = page.ImageObjects[j];
string text = obj.RecognizedText;
if (text == "ISBN")
{
// We want to use this text for reference. Create a reference element and define its geometry on the current page.
ITrainingField isbnTag = newDefinition.Fields.AddNew("ISBNTag", TrainingFieldTypeEnum.TFT_ReferenceText);
page.SetFieldBlock(isbnTag, obj.Region);
break;
}
}
// We assume that we have succeeded in defining two items
assert(newDefinition.Fields.Count == 2);
}
// After the user has defined the layout (or verified the automatically computed layout for subsequent pages), he must submit
// the result for training. In this sample we assume that the computed layout is always correct, so just mark the page as verified
// and ready for training.
trainingBatch.SubmitPageForTraining(page);
// Fetch the next page that requires attention. The method will return null when all the pages have been verified and submitted for training.
page = trainingBatch.PrepareNextPageNotSubmittedForTraining();
traceEnd("OK");
}
traceEnd("OK");
trace("Export to AFL...");
newDefinition.ExportToAFL(batchFolder + "\\NewTemplate.afl");
} finally {
trainingBatch.Close();
}
trace("Create document definition.");
IDocumentDefinition newDocumentDefinition = engine.CreateDocumentDefinitionFromAFL(batchFolder + "\\NewTemplate.afl", "English");
trace("Use the new document definition with FlexiCapture Processor.");
CheckTrainedDocumentDefinition(engine, newDocumentDefinition, rootFolder);
}
