The best classification results are achieved if the OCR font is trained using real data from the target application. This can however be time-consuming. To speed up the creation of a large number of different samples per symbol, it is possible to vary existing samples.
To add variations for certain samples, select either
Then open the Generate Sample Variations dialog either via the corresponding toolbar button or via the Edit menu.
This dialog allows you to select several types of variations, depending on what is required for the application. In general, we recommend using as many samples as possible. So, if you are in doubt whether a variation type applies or not, you should select it.
The following types of variations can be selected:
The OK button starts the generation of the new samples. Note that the generation might take some time if lots of variations are selected for a large number of samples. A progress bar shows how far the generation has proceeded. The generated samples can subsequently be viewed and edited in the Sample Inspection Window where they are marked with true in the Generated column.
Note that once the training file is saved, the generated samples cannot be recognized as generated ones any more and therefore the Generated column will show false even for a sample that has been generated.