I would suggest you fire up Google and look through the various OCR solutions that are out there and the code that people are using to work with it.
Google Search: C# OCR[
^]
Once you get that working and can start pulling the data from the images... your next task is going to be converting it into a data model so that you can verify all of this data makes senses; aka validation.
The last step is going to be taking the data model and submit it to your ERP.