(Archived) Optimum picture resolution for Evernote OCR?

[steveberl 2]

What is the optimum relationship between font size and picture resolution for the evernote OCR functionality?

I would imagine that there is some range of "pixels per character" which is large enough for reliable character recognition, but not too big so I waste a lot of bandwidth and overrun my monthly quota.

Is it reasonable to do character recognition if the characters are 12 pixels high? How about 10? 8?

How badly does the reliability fall off as the pixel height of the characters goes down?

Just wondering,

-steve

[jbenson2 2,149]

I don't know is this an accurate test, but I gave it a shot anyways. I added the names of some Minnesota counties to the photo.

I imported this photo into my Evernote notebook.

Within 2 minutes of syncing, I was able to search and find all county names from Dahkota 12 down to Kittson 4.

It did not find Pembina 2, but to be honest, I can't even read it myself.

Here is the link to the photo: http://bit.ly/3jUS16

[steveberl 2]

Cool.

I zoomed in on your image and it appears that the Kittson 4 text is 9 pixels high and the Pembina 2 is only 5 pixels high.

So it seems like a character height of 9 or 10 pixels might be as small as one would want to go for accurate, reliable OCR?

[engberg 89]

There are a million variables around things like clarity, font type, etc, but I've definitely seen good results in this range. Basically, if it uses a clear dark text on a light background with good contrast, and it's easy for me to read as a human, the image processing does a good job. If I have to squint at all, we'll probably have a hard time.