OCRExtractRelative captures data correctly but information is truncated

[Wilson_Rocha]

Hi,

I am facing a problem with OCRExtractRelative that truncates a name. It works fine most of the times but for two specific cases, same names that are read correctly a dozen times, are truncated. The green box has already been increased and the pink box is large enough to capture, once the words are captured correctly for all other cases.

Last screenhots show that the images are fully captured. Variable contents, however, are truncated: SUPORTE COMPARTILHADO shows as SUPORTE COMPARTILHAD and INTERLIGACOES SUBMARINAS shows as INTERLIGACOES SUBMARI.

Please find attached a sreenshot of the page, relative image utilized and the last screenshots of the two cases.

Thanks in advance.

[ulrich]

I found that with a little bit more white space around the text the OCR API is usually better. Try with a bigger pink box. For that, you can also shift the green box a bit more to the left. Maybe like this:

Are you already using OCR Engine2? It is better for Western text OCR.

[Wilson_Rocha]

Hi Ulrich,
Yes, I am using OCR 2. I tried some options with more space around the text (before, above and below), but, unfortunately, the two specific cases still return truncated information.
Anyway, thank you again for sharing your knowledge.
Best regards

[admin]

Can you share the two _lastscreenshot.png of these two cases?

[Wilson_Rocha]

HI admin,
The two _lastscreenshot.png for these cases are the last two files attached above. They seem fine, the correct infomation is there and are: SUPORTE COMPARTILHADO and INTERLIGACOES SUBMARINAS. For some reason, the variable that contais the information is truncated.
Thanks for your attention.

[admin]

For me it works if I add more white space around it. So your version fails, but this one works:

Also, does it have to have the blue background? If you can avoid that, it would be easier for OCR.

Test:

[TechSupport]

We made a small tutorial for improving the OCR result:

OCRExtractRelative: A larger pink box OCR area is usually better

We found that a typical root cause for bad/wrong OCR results is that the pink box (that marks the area for text recognition) is made too small. This is especially true for number OCR. Solution: Add more white space around the text And (for number OCR) to include some additional text (e. g. the label). If you do not need the text part, you can remove it after the OCR by splitting the OCR result string.