Grooper 21.00.0082 is available as of 12-12-2023! Check the Downloads Discussion for the release notes and to get the latest version.
Grooper 23.00.0042 is available as of 03-22-2024! Check the Downloads Discussion for the release notes and to get the latest version.
Grooper 23.1.0018 is available as of 04-15-2024! Check the Downloads Discussion for the release notes and to get the latest version.
Options
Correcting OCR Data after Form Identification
hjanum
Posts: 110 ✭✭
We handle multiple forms in our solution. After the documents have been processed by Grooper they proceed to a downstream process where they are digitally signed. The digital signature solution places visible PDF signature(s) based on the positions of certain words and phrases in the OCR text of the document. It is therefore crucial that certain words/phrases are (mostly) correct in the document. We do have some fuzzy matching, but not as advanced as in Grooper.
We do use a multi-engine OCR approach in Grooper that uses an extractor to fix some of the labels, but that will not always work for us.
Take the following example.
Say we include the string "Form 118" in the lexicon and set a match percentage for say 90%. This will fix the string "Form l18" where the first one has been read as a the lower case letter L. We have one form 118 document where the OCR result is "Form 116". I could allow replacement of "6" with an "8" using weights and that would fix that particular document. The problem is that we also have a form 116 that would say "Form 116", so in that case we would not want to correct the "6" to an "8".
We do however have a well tuned form identification in Grooper using weights. It works VERY well and it has not mis-identified a form yet. So, later in the process we know for sure we are dealing with a form 118, so in that case it is perfectly fine to replace the "6" with an "8" in the OCR results.
There is a "Correct" type process that can be used to correct OCR results based on replacement/lexicons like the multi-OCR engine approach?
I see the Correct process, but I don't see the option to use extractors in that one. I would also need somehow to pick different lexicons based on document type, as the corrections I would want to make would depend on the form type.
Is this possible?
We do use a multi-engine OCR approach in Grooper that uses an extractor to fix some of the labels, but that will not always work for us.
Take the following example.
Say we include the string "Form 118" in the lexicon and set a match percentage for say 90%. This will fix the string "Form l18" where the first one has been read as a the lower case letter L. We have one form 118 document where the OCR result is "Form 116". I could allow replacement of "6" with an "8" using weights and that would fix that particular document. The problem is that we also have a form 116 that would say "Form 116", so in that case we would not want to correct the "6" to an "8".
We do however have a well tuned form identification in Grooper using weights. It works VERY well and it has not mis-identified a form yet. So, later in the process we know for sure we are dealing with a form 118, so in that case it is perfectly fine to replace the "6" with an "8" in the OCR results.
There is a "Correct" type process that can be used to correct OCR results based on replacement/lexicons like the multi-OCR engine approach?
I see the Correct process, but I don't see the option to use extractors in that one. I would also need somehow to pick different lexicons based on document type, as the corrections I would want to make would depend on the form type.
Is this possible?
0
Answers
There is also a removal extractor you can use within the Correct activity. It's job is simply to find text in the OCR results and delete it.
Product Manager
mharrison@bisok.com
The question was mostly doing the correction specific to the identified form type. Thinking some more on this I think I can do it with the "Correct" activity if I put a "Should Submit Expression" that checks the form type, so for Form A I run one type of correction and for Form B I run another type of correction.
I would then have one Correct activity per form type each with its own Correction Extractor.