Issue:

Sample documents are all poorly scanned with very light text. This results in poor whole page HOCR results for lucene classification. As a results, Search Classification is inaccurate.

Solution:

1. It is best to provide the best possible sample training documents. If this can’t be done, try editing the Batch Class FPR.rsp file in Recostar Design Studio to improve HOCR results.

Add BinaryImageSequence->RepairMatrix 

2. Re-Learn Files after saving FPR.rsp file.

Was this article helpful to you?

Walter Lee

Comments are closed.