OCR Failure

OCR_IMAGE

I am getting this error

"WARNING: INVALID RESOLUTION 0 DPI. USING 70 INSTEAD."

occasionally when OCR ing an image such as the one above. I am using the Latin language.

Here's what I get taking a screen shot of just the image in question:

Warning: Invalid resolution 0 dpi. Using 70 instead.

But when I take a screen shot of a slightly larger area:

OCR Failure
© wsdeena while since we've s
e Ryan_P
= NNDM

Thanks for testing. Yes is wierd. I have tried adding a margin to the capture and setting the DPI from insifr keybpard maestro, to no effect. Just shifting the capture location by a few pixels often is enough though. I can't increase the size of the screen capture itself as it makes it difficult to seperate out the symbol I am trying to OCR from other text.

Worked some more on this. I believe the DPI error message is just indicating that the OCR failed.

An interesting article on Tesserect failures

. It seems Tesserect performs best with text formatted in a document structure, and less well in my use case. They had some success performing binarization on image. Not quite sure the easiest way to go about doing this.

2 Likes

Tried adding margin but got same error. Also tried resizing, setting DPI but no dice

I went through all of the article and even followed the linked sources they cited but they didn't offer any working solution which is the domain of any developer including KM's one, anyway.
Also, since it's unreliable I see no benefits of having it in the first place because you may plug in any dedicated software - both proprietary and open source.

I found this topic after getting this same error today. I decided to run an Automator workflow containing a Pixelmator Color Adjustment action over my image before attempting OCR and I'm now getting results. Something to consider if anyone ever comes looking in the future for ideas.

1 Like

That sounds great! Would you mind sharing your Automator Workflow? Probably need to upload it as a zip file.

Thanks.

I'm getting this, which wouldn't bother me but it takes over my clipboard every now and then. I suppose the solution is to make sure the OCR result goes to a variable instead of the clipboard.

for OCR on the fly, there are two great options, TextSniper and OCROwl