Entropy said:
Or, select the offending character and get a list of the most probable matches and choose the right one? So far, it's pretty obvious to me when the OCR engine gets it right, even though in the demo I never see it turn to blue.
That's fine for single-character errors, but maybe 25% of the time you'll get a page of total gibberish and have to re-take the whole thing; it's usually obvious right away at least, but it can be very annoying if you've already moved on from the thing you took the picture of and have to go back / frame it up again.
Entropy said:
Unless there's so much green gibberish that you can't see the original character. I'd like to be able to divide the screen so i see the raw image and the overlay in separate panes.
You can turn off the "green gibberish" with the "hide unused chars" option in Settings / OCR. Though an option to move the overlay might make sense too... I'm not quite sure where we could fit it on that already-overcrowded screen, though.
Entropy said:
Will I be able to download it after I buy it on my phone? Much easier to test still image rec on an iPad.
Unfortunately no - that does work on the iPod Touch (though you have to load the OCR data files manually and we haven't posted the link yet) but it won't even let you in on the iPad; as I said, we didn't want to spend time designing / testing an iPad interface for something we weren't yet selling on iPad.
dustpuppy said:
i think for the still image mode, you should let the user pan and zoom the target image, which would mimic moving the phone. that way, i can take a nice crisp picture of a sign with perfect focus, then perform character recognition easily with having to keep my hands steady
That's a possibility, but it doesn't really get around the problem of the recognizer not seeing an image correctly - the reason "live" panning works is that it's sifting through hundreds of frames of video, each with a slightly different angle / focus / lighting level / etc, while even a well-shot picture of a sign might not be recognized correctly. So it's requiring extra work on the part of the user (moving around the box) without the attendant improvement in accuracy that you get from live input.
We're actually leaning more towards something like the document reader, but with a photo - tap on a character in an image and get a popup definition bubble just like in the document reader. A key advantage to that is that it keeps the actual image right in front of you, so you can see exactly where a particular word appeared on a page and easily go back and forth between the original document and the onscreen version without having trouble keeping track of your location in it.
Luke said:
The OCR is great, Mike. Thanks so much for working on the feature, it's hugely to so many of us. I'm fighting with iTunes to actually be able to purchase it (I'm in China but my Apple Store account is in the US…I think that complicates things), but on my iPhone 4 the OCR in demo mode has been impressive in its ability to correctly recognize characters. It will be a real life-saver for me here in Beijing.
Thank you! If you're getting rejected credit card errors, the easiest way to clear them up is to contact iTunes support and ask them to unblock your account (
http://www.apple.com/support/itunes/contact.html) - they get a TON of email like this and can usually fix the problem very quickly.
elipio said:
Mike, the OCR is out this planet! The silver detail in the icon is superb too... the coolest icon on the front page of my iPhone 4.
Heh, thanks - actually that was inspired by a user who wrote to say that he loved the product except for the icon and the splash screen. (we're still working on the latter, though it's about to become less of an issue now that OS4 / multitasking are finally arriving on iPad) But the old icon didn't even have correct drop shadowing, so a change was certainly overdue, and it was perfect timing since our icon's temporarily showing up in a bunch of places in iTunes that it doesn't normally (the iTunes-wide top grossing list, e.g. - we made more money yesterday than Flight Control, though that's considerably less of an accomplishment than it would have been a year ago
).
John said:
Does the OCR depend on where in the box the character is ? I could get the OCR to easily recognise two characters earlier (displayed on my computer LCD screen) but as soon as I shifted the iPhone so that the left side of the box only enclosed the second character, the OCR had trouble recognising it.
It can, for two reasons: #1, with only a single character to look at the recognizer might be having difficulty figuring out the correct font height (whether it's one character or two, e.g.) - usually not a problem but it occasionally does happen with certain characters - and #2, if the box includes a little bit of the previous character that might be confusing it. A good workaround when this happens is to point at the full word, tap the pause button, tap the > button, then tap on the character in the separate dictionary entry screen to look it up.
