We are OCRing hand-written numerical dates (e.g. 12/24/24) and Document Intelligence is doing a great job on all numbers except ones. It doesn't get them wrong - it doesn't "see" them at all.
We have stripped the form of other vertical lines and slashes in hopes that would help. It has not.
For example, we have separated the dates components into distinct areas on the form, with no boxes or front-slashes needed to separate the month, day, and year, and Doc Intel is still not seeing ones.
Any suggestions on best practice for this?
As per Read OCR Handwritten text extraction, I found few checks you can do to improve Azure Document Intelligence's recognition of handwritten numbers. Apart from this, ensure scanned documents have high resolution, clear contrast, and legible handwriting to enhance text clarity. You can also use confidence scores returned by the model to flag low-confidence predictions for review or correction.
Reference-