Read numbers on a metal plate

[PKD]

Hi,

I have images where numbers are curved on a metal base. The standard tesserart approach is not returning the numbers. Tried applying relevant page segmentation method - but could not get any accurate reading of the text.

Any guidance \ indicator on how to approach such a problem will be very helpful.

Regards.

[elton fernando]

Hi,

As you said you already tried preprocessing and it didn't work, you should train the tesseract, use the version 3 method, where training is done by individual character. 

version 4 uses blocks (words) doesn't usually have good results for this case.

--
You received this message because you are subscribed to the Google Groups "tesseract-ocr" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tesseract-oc...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/49fda3a9-5ab3-4a70-b881-7681180351ecn%40googlegroups.com.

[Zdenko Podobny]

There are 2 steps:

1. text detection in the photo - IMHO this is the most difficult part. Try to search for "text detection in natural scenes".
2. text area processing for OCR
3. OCR.

E.g. with this  image you can get quite good result with tesseract:

tesseract engraving.png - --psm 6 --oem 1
S601 25835

As you see only problem is the letter "G" detected as "6". Maybe better processing (step 2.) can fix it easily:

Try to search this forum for "embossed", "engraved", swt.

IMO there are several tutorials for OCR embossed credit cards, that could help you. e.g. https://www.pyimagesearch.com/2017/07/17/credit-card-ocr-with-opencv-and-python/

or https://stackoverflow.com/questions/9398374/prepare-complex-image-for-ocr

Zdenko

po 7. 2. 2022 o 13:29 PKD <pran...@gmail.com> napísal(a):

--