Train Character dialog box
The environment window displays the image of the character(s) to be trained along with some context. A highlight shows which character(s) will be trained.
The zoom window displays a pixel-by-pixel image of the character shape(s) to be trained. Blue pixels show the active shape, yellow pixels the neighboring but excluded shapes. Always the active (blue) shapes will be trained.
Broken shapes (e.g. a broken n recognized as rr) or touching shapes are editable. You can join fragments. The image handled by the program as one character (displayed in blue) can be widened but not narrowed down.
To join a missing fragment to the desired character shape, click on it. The clicked fragment will change from yellow to blue indicating that it will be included in training.
See the example below: a broken f is displayed in the zoom window.
|
The blue image (active shape) may only be part of a character, as shown to the left. Click on the second yellow part to join it to the active shape. |
After clicking on the yellow part it will become blue, so the whole f shape will become active as shown to the right. |
|
If you previously joined fragments in error, you can undo this. Click on a joined fragment to detach it. The clicked part will change from blue to yellow indicating that this fragment will be excluded from training.
Suggestion displays the OCR solution offered by the program for the character(s) highlighted in the environment window and displayed in blue in the zoom window.
Enter one or more characters in the ‘Correct’ box as the correct interpretation of the character shape(s) displayed in blue in the zoom window. Typically you train one character, but it is possible to train up to 5 characters, within one word. This lets you train ligatures (e.g. fi, ff) and wrongly segmented words (e.g. ‘run‘ was misrecognized as ‘nm‘).
Click to open the Character Map.
Click Train to accept your training so it can influence recognition in the document. It will be provisionally added to a training file, if one is loaded. If not, it is stored along with other training (manual and IntelliTrain) and you can save it to a training file if you wish.
If characters with similar shapes are found on the same page, the Check Training dialog box appears. If nothing appears after clicking Train, then no other corrections are proposed, all you did was correct the single occurrence.
-
Do not use training with Japanese, Chinese or Korean.
//