Improving handwriting OCR on cloud vision?

Question

I was trying to analyze the following image, and convert it to text, using Google Cloud Vision

I realize that it has low quality, and that some words aren't exactly uniquely identifiable. My current strategy was to threshold and splice into sections using numpy and opencv. Then, for each section I use the document_text_detection function to get the results for each section of the sentence above. However, this gives extremely poor quality. I am not looking for 100% accuracy, but just something that looks fairly close to what is written in this line. Any suggestions to improve the result? Are there other tools than GCV that could be useful?

I did. I used Google Cloud Vision, but it doesn't give great results. — Drew Brady, Feb 02 '22 at 18:47
Can I remove the solid line under the characters manually in my algorithm? — Prefect, Feb 03 '22 at 13:00

score 4 · Answer 1 · answered Feb 07 '22 at 23:12

4

Vision API has a new OCR model available for opt-in. You can try it out by specifying the model "builtin/latest".

More info here: https://cloud.google.com/vision/docs/release-notes#January_21_2022

answered Feb 07 '22 at 23:12

Brendan

1,017
5
7

score 0 · Answer 2 · answered Feb 02 '22 at 13:48

0

You could give Transkribus a try, which has been reported to provide good OCR results on hand-written historical documents.

answered Feb 02 '22 at 13:48

mara004

1,435
11
24

Improving handwriting OCR on cloud vision?

2 Answers2