How to train the Receipt Capture engine?

  • Last Post 14 July 2017
José da Silva posted this 13 July 2017


From reading the Receipt Capture SDK, I got the notion that the engine can be trained to read new fonts, new languages and new receipt formats:

ReceiptEtalon Object (IReceiptEtalon Interface)

This object provides access to the receipt model.

The receipt model is used to identify the vendors which can produce the receipts you process and to assign the corresponding purchase type to each vendor. Pretrained receipt models are available for some countries, and you can improve them by adding the typical receipts which your application is expected to process.

I haven't found on the documentation how to perform this training for new receipt models, only for "commodity" objects.

Is it possible to do it myself, or do I need to provide receipts to ABBYY to train new models? If it is possible, where can I find a code sample on how to do it?


Also, can the OCR engine be trained to read new typefaces and new languages?

Oksana Serdyuk posted this 14 July 2017

Please review the Working with the receipt model section of the Recognizing Receipts article (Developer's Help → Guided Tour → Recognizing Receipts).

As far as adding new typefaces and languages is concerned it should be done in the OCR technology and it can be implemented only on ABBYY side.