OCR doesn't identify anything in this image

[JD]

I've used the Pleco OCR quite a bit and love it.

However, today I ran across an image that surprisingly gave me nothing at all.

It's a screenshot on my iPad from a game called "Crossy Road" with the language setting in the game set to Chinese.

Is the problem something to do with the white text on orange/red backgrounds? I tried flipping the setting for the white/black detection, but it doesn't help.

I literally get ZERO identification of any characters, unless I zoom in on a single character, in which case, I get something completely incorrect.

JD

EDIT: if I zoom the picture first and then put the bounding box around the small pieces of text, it does at least try to identify something...but it is still WAY off.

 

[mikelove]

That's odd, I just pulled up this attachment in OCR here and it worked perfectly - is this the original image or was it converted? Any chance the vertical/horizontal text toggle might be configured incorrectly?

 

[JD]

Extremely bizarre....I downloaded the copy from the forums and it worked just fine. However, the original doesn't. I cleared all the old images from the Still OCR and it still does not work for the original.

In fact, I just did an iPad screenshot of Pleco NOT working on the original, and then imported that into the Still OCR and the iPad-copy-of-the-iPad-original works just fine.

Here's what I see when I try to import the original attached (just notice it has a question mark on the orientation whereas it does not on the forum copy...another oddity).

EDIT: I toggled both the horizontal/vertical and the black/white on the original, but it did not change things.

EDIT 2: The image was taken by doing the "hold home and top-switch" on the iPad Air and then selecting the Still OCR in Pleco and grabbing the image from the Photo Library. hence, nothing tricky. If there was any image conversion, it would only be in the Upload to the website in the background. But that doesn't explain why the screenshot from inside Pleco would work.

 

[JD]

Ok, I zipped the original image file on my iPad and then gmailed it to myself so that I could use my PC to upload the zip to the forum. (I don't appear to be able to upload .zips from my iPad)

In parallel, I also emailed the image from gmail back to another email account on my iPad.

After unzipping this image on my iPad, I still see that it does NOT recognize the characters when I import that image into Pleco's Still OCR from the Photo Library.

Take a look and see if you see the issue with this attachment.

JD

 

[mikelove]

Try turning on "Shrink Big Images" in Settings / OCR / Still / Block Recognizer - does that help any?

 

[JD]

Yes, that does appear to be a workaround when I turn on that setting. However, the original image, at 251KB, is smaller than other working images (e.g. an image of 275KB works fine without telling Pleco to shrink images).

What is that setting doing?

EDIT: and the image size stays the same when it's uploaded to the forum website and then downloaded again, but works in the latter case, but not the former.

 

[mikelove]

It's literally just resizing the image internally to 50% - we do that automatically in some cases but apparently the filter we use for that was triggering intermittently for you (may have been right on the edge of a threshold so that even a minor format change would push it over). Our OCR engine has a limited range of character sizes it recognizes well (necessary for performance reasons and to avoid finding even more random giant characters all over the place) so resizing helps us find characters reliably in more images.