Please note: This documentation concerns to the prerequisites and specifics wich aren't described comprehensively in the Abbyy Recognition Servers manual, regarding to the magazine "Daheim" and comparable digitized media.
These settings can be accepted as provided. It saves time-consuming troubleshooting.
The Recognition Server manual is related to a Windows environment and Windows network shares. If the OCR results should be saved onto a network share, you need to select a user with privileges on accessing the network. The local system account has comprehensive privileges on the local machine, but none on the network. It's advisable to choose the network service account. In our case we had Novell domain and Novell network shares. These seem to be incompatible in this case. The error message of Recognition Server was cannot access drive xxx (the network share). Thus we saved the OCR results on the local server and copied them onto a network share by using a synchronizing programm.
It's advisable to create a folder structure like the following:
These users have to be an account with password on the server
Mount the higher folder, e.g.Daheim as network share on computers installed with Verification Station.
This option arranges the picture on the basis of straight lines. If they are like in this example, a trapeze-shaped flag, the result looks like this:
That's why you should deactivate the checkbox Entzerren/dewarping
Documents with simple text formatting, e.g. continuous text, you may choose "Keine Überprüfung/No check".
Documents with more sophisticated formatting the option "Alle Dokumente überprüfen/Check all documents" should be activated.
If Abbyy Recognition Server doesn't recognize some characters, a training file can be created in FineReader, exported as .fbt-file and imported into Recognition Server. A demo version of FineReader can be downloaded on the Abbyy-homepage.
In order to create a training file, e.g. for the Gothic type, set the checkbox "Use training in order to recognize new characters and ligatures". Now click on "read page". A dialog field opens.
In this case the letter "M" wasn't recognized properly. Extend the box by using the opening double arrow >> until the letter is completely covered. Following click on "Training". If there is more than one letter boxed, execpt on ligatures, you can shrink the box by using the closing double arrow << until the single letter is covered on it's own.
The training can be finished at anytime. When you are finished, click on Close
The trained letters can be reviewed by
Their properties like bold or italic style can be set as well.
In order to check it the accuracy of the OCR you can OCR some documents with and without your custom training file.
The training file can be exported to Recognition Server by clicking on
and then add this file like described in the workflow settings.
If there's a text box without page numbers, even though it is enclosed, maybe the font color is the same like the background color. Click on the double arrow like shown in the picture to open the menu, click into the color box and choose black.
Everytime a new text box is drawn, the print type is set to default, which only includes "normal" print types. If there is gothic to be recognized, the check on Gothic has to be set.
After this you need to check the reading order and correct it.
The verification station has a built-in spell check, that can be invoked by clicking the corresponding button. As the spell check is quite faulty, sooner or later error messages will appear (see Error messages). Alternatively, a regular text-file with UTF-16 formatting can act as dictionary when you apply it to the workflow settings 'It's added via 2nd tab -> Verarbeitung/Processing. -> interner Link zu "Tab Workflow -> Workfloweinstellungen/Workflow settings
occur when:
These are bugs in the program.