(Archived) V disappointed, OCR not working on wine label at all

[karlengel 0]

My first test of EN Mobile is a big disappointment. Uploaded a wine label photo from my camera phone using the Snaphot button (HTC Tytn II). It's in focus, quite well-lit, good contrast, image size shows up on desktop version as 291KB. Not a single word of the label is returned in searches. Have sync'ed several times over several hours. Fair enough if it was a dodgy picture, or just the most curved words that aren't recognised, but some words are quite straight in the middle. Here's the original image:

http://www.flickr.com/photos/39354678@N07/?saved=1

Am I being unreasonable in my expectations? I was hoping to use EN frequently for this sort of thing.

[engberg 89]

Can you search for "product" or "Australia" if you log into your account on the web and search there?

[karlengel 0]

Neither "Product" nor "Australia" return this note in searches from the web ap. I'd be interested to see if you get any different outcome loading the same shot in. Thanks for the prompt reply by the way.

[engberg 89]

flickr doesn't give access to the same raw image that you were putting in. I could look at it in a little more detail if you provide your Evernote username or else a URL to the raw JPEG from your camera.

Alternately, you could put the note in a notebook that is public and give the public notebook URL.

[karlengel 0]

Username is bwva

[engberg 89]

That image wasn't processed because you added it with a file suffix of ".jpg" (IMAGE_257.jpg) and the contents of the file are actually PNG. My guess is that you put a real JPG file into Flickr, and then grabbed a PNG image back from their web site with the wrong name. If you put the original JPG from your camera into Evernote instead, you should get decent results.

We have code on the server that tries to check incoming image files to see whether they are valid and within our acceptable bounds for sizes before we run them through all of the image processing libraries. In this case, we saw that the file claimed to be JPG so we ran it through the list of JPG checks, and it failed immediately because it wasn't a JPG.

In theory, we could try to get a little smarter by looking into the contents of the file to see whether the client lied about the type, but this isn't how it works today.

[karlengel 0]

Here's my guess of what happened, but it raises a new issue:

I uploaded the shot direct from my camera via EN Mobile's Snapshot feature. It should have hit EN as a jpeg. It was taken in Portrait, so it came up in EN sideways. I think I used the Rotate editing tool in EN desktop version to make it vertical. Did this change its format to PNG before OCR had been done? I think I noticed it was in PNG format when I went to export it to Flickr, but assumed this was the format EN uses to export all picture images so as to avoid recompression. Does EN change jpegs to PNG during editing functions? Was it simply that I didn't wait long enough (for OCR) before editing? In any case, do images which have been processed as jpegs by OCR still return correct OCR search results if they are subsequently edited and turned into PNGs?

PS

Well now I'm confused. Just rotated another camera phone image ("OCR Test") in EN and it offers to save as a jpeg, not PNG. Went to the Flickr site, and the original image offers save only as jpeg. Don't know how the Wine Shot image became a PNG. Unless rotating transforms it internally but saves still occur as jpegs.

[engberg 89]

Ah, I think I see that you're right. The PNG conversion using our built-in image manipulation dialog in our Win32 client is a bug. We'll take a look at this, but for now I'd recommend using the right-click "Edit" option instead to open the image in your favorite image editor, then Save it back into Evernote.

Sorry about that, thanks for the report.

[karlengel 0]

And now the original problematic shot "Wine Shot" is OCRing correctly. Did you change it back to a jpeg, or did it somehow fix itself?

Weird, it's OCRing correctly on my desktop version, not on the web.

[karlengel 0]

Two images I've recently uploaded as jpegs - and have not manipulated - now offer PNG as the Save format. These are OCRing correctly. The same images which I had manipulated within EN offer jpg as the save format. These are OCRing correctly on the desktop but not on the web. Other images I had previously uploaded as jpegs, and not manipulated, offer jpg as the Save format. I don't really care as long as it all works, but this seems strange.