Form Recogniser form labelling tool - offset tables

Michael Douglas 21 Reputation points
Dec 15, 2021, 5:22 PM

I have looked at the table function on fott but I cannot see how I can deal with the sort of table below which can extend onto multiple pages.
The "column" headers are not really columns and the "rows" are not really rows, but obviously it is easily understood by a human.

I also looked on AI Build but I assume it is using the same background tools.

Can you advise the best way to tag and train this sort of table please?

Many Thanks
Michael

158011-image.png

Azure AI Document Intelligence
Azure AI Document Intelligence
An Azure service that turns documents into usable data. Previously known as Azure Form Recognizer.
1,787 questions
Azure AI services
Azure AI services
A group of Azure services, SDKs, and APIs designed to make apps more intelligent, engaging, and discoverable.
2,971 questions
{count} votes

Accepted answer
  1. YutongTie-MSFT 53,341 Reputation points
    Jan 4, 2022, 11:44 PM

    @Michael Douglas

    Hello Michael, I did some research on my end and I think it's kindly working for me.

    My solution here is custom model with V3.0 Form Recognizer in Form Recognizer Studio. Please check the link to see what is custom studio and what is FR studio.

    What I have tried is, I created a table in FR studio and labeled related data as below:
    162334-image.png

    To transform those data to a form as the right.

    Since the custom model function need at least 5 similar form for training but I only have one, my result is not kind of accurate, but it's close to what I expect to as below:

    ![162313-image.png]3

    I think this should be a good solution for your business. But one thing I should mention is, we are not supporting multi page tables, which means every item section should be in same page.

    Hope this will help. Please let us know if any further queries.


    • Please don't forget to click on 130616-image.png or upvote 130671-image.png button whenever the information provided helps you. Original posters help the community find answers faster by identifying the correct answer. Here is how
    • Want a reminder to come back and check responses? Here is how to subscribe to a notification
    • If you are interested in joining the VM program and help shape the future of Q&A: Here is how you can be part of Q&A Volunteer Moderators
    1 person found this answer helpful.
    0 comments No comments

4 additional answers

Sort by: Most helpful
  1. Michael Douglas 21 Reputation points
    Dec 29, 2021, 12:12 PM

    Many thanks Yutong - really keen to get something on this :-)


  2. Michael Douglas 21 Reputation points
    Jan 4, 2022, 9:10 AM

    Hi Yutong

    Yes you have understood the table correctly :-)

    Basically the top 2 sections (left and right) are the legend for each Item. There can be up to 99 items per document.

    Best Regards
    Michael

    0 comments No comments

  3. Michael Douglas 21 Reputation points
    Jan 6, 2022, 2:42 PM

    Hi Yutong

    Many thanks for this - I can see the process now.
    Unfortunately many of these are on multiple pages so I will look at alternative solutions from the other major providers.

    Appreciate your efforts :-)

    Best Regards
    Michael


  4. Michael Douglas 21 Reputation points
    Jan 18, 2022, 9:44 AM

    Hi Yutong

    The barcode is actually the MRN number which is stated in text nearby. However it is useful to read the barcode as the MRN number is not always in the same place.

    Just for your information, this "EAD" extraction requirement is part of a larger "Blockchain Europe" project which I am involved in with the Fraunhofer Institute. They are also looking to read EAD's and will start soon to do their own work. But I would prefer to use an existing major solution such as Forms Designer, so any development you do is really useful for us.

    Kind Regards
    Michael

    0 comments No comments

Your answer

Answers can be marked as Accepted Answers by the question author, which helps users to know the answer solved the author's problem.