Train Font
The Train Font tab is used to train, view, rename and remove characters. The tab is divided into two groups, Characters, where the trained fonts are managed, and Training, where the character training parameters are defined.
Note:
- The controls on this tab are not linked to the Train Mode, Train String and Train Font parameters on the General tab.
- While in the Train Font tab, the ROI can be adjusted (the Edit Graphic button is enabled) and the segmentation graphics (character regions) will be updated as the ROI is modified.
- Multiple-byte character names (e.g. Kanji) are not supported. Multiple-byte characters can be trained, but the name given to the character must be a single byte name.
Characters Box Controls
After training, each trained character will be visible in the tree, where the character will assigned a folder and label, and grouped together (if the characters have matching labels).
- Select the root Font folder to display all of the characters in the font in the panel on the right-side. The characters will be displayed as icons with a label underneath.
- Select a character folder to display all of the trained instances of that character in the panel on the right-side. The characters will be displayed as icons with a label underneath.
- Select a trained instance of a character to display that character, zoomed to fit the panel on the right-side.
Tip: Providing the OCRMax function with multiple training instances of each character to be read during run-time allows the function to better discriminate between similar character types. However, training multiple instances of the same characters has advantages and disadvantages.
The advantage to training multiple instances is that better classification performance can be expected, especially when reading text that could easily be misclassified (e.g. 8 and B). The disadvantages of training multiple instances are: a decrease in the function's execution time; possible decreases in the confidence scores, because multiple training instances should not only improve the highest score, but also improve the second-highest score, as well; and possibly training bad characters.
The advantage to training multiple instances is that better classification performance can be expected, especially when reading text that could easily be misclassified (e.g. 8 and B). The disadvantages of training multiple instances are: a decrease in the function's execution time; possible decreases in the confidence scores, because multiple training instances should not only improve the highest score, but also improve the second-highest score, as well; and possibly training bad characters.
- Import Font: Launches the Open dialog, where a font, saved as an OCRMax font file (*.omf), may be imported.
- Export Font: Launches the Save As dialog, where a font can be saved as an OCRMax font file (*.omf).
- Rename: Press after selecting an item to rename it. Any element in the tree or in the list panel on the right-side can be renamed.
Note: Renaming a group of characters will assign a new label to all of the characters in the group. Renaming a single character will assign a new label to that single, selected character; the character will be moved to a new or different group in the tree.
- Delete: Press after selecting an item to remove it from the font. Any element in the tree or in the list panel on the right-side can be deleted.
Note: Deleting the Font folder will clear the font; all characters will be deleted. Deleting a group of characters will delete all of the characters whose label matched the deleted group's label.
Training Box
Before characters can be trained, they must be correctly segmented.
- Add All Characters to Font: Specifies that all of the characters in the ROI will be trained. The characters expected to be trained are entered into the Train String text entry box, before the Train button is pressed. The number of characters in the Train String text entry box must match the number of segmented characters.
- Add New Characters to Font: Specifies that only new characters in the ROI will be trained. The characters expected to be trained are entered into the Train String text entry box, before the Train button is pressed. The number of characters in the Train String text entry box must match the number of segmented characters.
- Add Individual Characters to Font: Specifies that specific characters in the ROI will be trained. When this option is selected, the Train String text entry box will be disabled, and the Train button will launch the Add Selected Characters to Font dialog. This dialog contains an unwrapped image of the ROI, with a label and text-entry boxes below each of the segmented characters. The label is the currently associated symbol for that character (a "?" denotes an unknown or untrained character). Below the label is a text-entry box. Enter a label for each segmented character in the text-entry box; leave the text-entry box empty to not retrain characters. Press the Train button to close the dialog.
- Train String: Specifies the text string to be trained. The number of characters in the Train String text entry box must match the number of segmented characters.
- Train Button: Activates training.