OCR!

mikelove

皇帝
Staff member
character said:
Happy Thanksgiving! Now to get that picked up by all the Apple-specific tech blogs.

Yeah, really have to thank Engadget for giving us two write-ups - even if we don't get picked up by anybody else as a result of that they generate a ton of hits on their own, almost single-handedly boosted us from #90 to #24 among top free reference downloads (though the sales bump wasn't quite as dramatic).
 
Congratulations on the Engadget love you've been getting! Must be very exciting.

I published a post on my website about Pleco OCR, more or less just expounding its virtues. I've been using it for over a week now and it's awesome. The post is here: www.chengduliving.com
 

telecafe

Member
Wow. I bought one and it's just great.
Anyway, sometimes OCR has missed to recognize letters with vivid-colored background, e.g. white-colored letters with red-colored background, and bold-typed letters.
Could you please improve the above cases? Thanks.

p.s. just curious, the above cases can be improved by Pleco? Or should be improved by Hanwang Technology which supplies Pleco's OCR core?
(I've just got to know that Hanwang supplied OCR besides HWR by your Legal Notices: http://www.pleco.com/ipmanual/copyright.html )
 

mikelove

皇帝
Staff member
justcharlie said:
I published a post on my website about Pleco OCR, more or less just expounding its virtues. I've been using it for over a week now and it's awesome. The post is here: www.chengduliving.com

Thanks! Really nice video - does a better job of showing it in a real environment than our official demo (which involved me rigging up a camera on a short tripod and trying very hard to keep everything in frame).

telecafe said:
Wow. I bought one and it's just great.
Anyway, sometimes OCR has missed to recognize letters with vivid-colored background, e.g. white-colored letters with red-colored background, and bold-typed letters.
Could you please improve the above cases? Thanks.

We already have a fix for background color / shadow problems, though it slows things down enough that we keep it off by default; go into Settings / OCR / Lookup Words and turn on the "Sauvola binarization" option. Might help with bold too, though often times those are screwed up by slightly-less-than-perfect camera focus (strokes blur together more easily in bold fonts); the iPhone camera's depth-of-field is terrible at short distances, you might want to look into getting an external lens of some kind to improve its focus.
 

adro

秀才
Thanks! Desktop version may actually be forthcoming, I'm thinking a $20 or $30 Chinese OCR app for Mac would be an interesting little way of experimenting with native desktop Mac development without committing to bringing the full version of Pleco to it. Along with perhaps a $10 to $20 drop-in replacement for Apple's trackpad handwriting recognizer that used our more accurate engine and two-finger clear gesture.

Hi Mike, I haven't been following the discussion carefully, so I might have misunderstood this message. But I'm just as interested in having Pleco dict on my computer as OCR!

Best,

adro
 

mikelove

皇帝
Staff member
adro said:
Hi Mike, I haven't been following the discussion carefully, so I might have misunderstood this message. But I'm just as interested in having Pleco dict on my computer as OCR!

That's a possibility too, but it's a lot more work than a self-contained little app like OCR so we're thinking small steps might be good here.


BTW, if anybody hasn't reviewed the latest version (2.2.1) yet we'd really appreciate it if you would - all of the attention from the Engadget post seems to have attracted a bunch of 1-star reviews (something like 8 in the last few hours, none of them with actual comments attached), and since it'll be a few weeks until 2.2.2 is out and our review average gets refreshed again, it'd be nice not to be stuck with a horrible rating until then. No need to type anything, just click on the the star rating in iTunes on your desktop or in App Store on your phone.
 

tanyahart

秀才
numble said:
Here's a feature request: A "scanning" mode.

Replace the definition window with a window displaying the current scanned words, the user hits "+" or "add/append" to append the current OCR selection to the scanned words. Maybe add quick keys for people to input punctuation marks, and some way to undo or backspace. The scanned items then are saved into a document or a pasteboard.

Good idea. I second.
 

tanyahart

秀才
Entropy said:
mikelove said:
Entropy said:
McCawley was back in print as of a couple of years ago. But the poor printing and the strange to Chinese-readers/writers organization--for example, 'tian' has 6 strokes in his scheme, and is probably found under "four-stroke enclosures"--make it hard to actually find anything when you need it.

Well that wouldn't be a problem in Pleco, but how do you find that the two compare content-wise?

Exactly--in Pleco you'd just search. McCawley seems pretty thorough for something dating to 1984, but apparently since then Sichuan cuisine has taken over most of China, and there's only a little of that in his book. I haven't gotten Hoenig yet, but my friend ordered two copies on Amazon which we should have in a few days. They currently have two in stock--order now! Presumably ezchinesey could give you an electronic form; I expect you'd have to work from paper for McCawley.

I'm going to PM you about this. I have a friend developing a specific menu dictionary, you should talk.
 

mikelove

皇帝
Staff member
tanyahart said:
numble wrote:
Here's a feature request: A "scanning" mode.

Replace the definition window with a window displaying the current scanned words, the user hits "+" or "add/append" to append the current OCR selection to the scanned words. Maybe add quick keys for people to input punctuation marks, and some way to undo or backspace. The scanned items then are saved into a document or a pasteboard.

Good idea. I second.

That one's already sort-of supported through "Capture Text" if you turn on Settings / OCR / Capture Text / Combine captured text; you also get a list of recently looked up words in the history screen.

tanyahart said:
I'm going to PM you about this. I have a friend developing a specific menu dictionary, you should talk.

PM received, I'll get back to you soon (digging out from under a pretty big pile of Stuff To Reply To right now).
 

character

状元
FYI, an OCR competitor of sorts: [deleted, in response to Mike's post below]

Based on the reviews it crashes all the time, but it's supposed to let you take a picture and look up characters.
 

mikelove

皇帝
Staff member
Please don't give that guy any more free publicity - I'm already ticked off at him for astroturfing about his Japanese app in the comments to our Engadget article. (half a dozen posts from three different newly-registered Twitter accounts)
 

tanyahart

秀才
That's the second (English) word I've learnt from this Forum in a week. First dumbphone, now astroturfing.

You don't have much to fear from the unmentionable app. It will annoy people all by itself. I tried it. It couldn't see the diacritics and read a different word. Same with Kanji. Very low accuracy rate.
 

kun4

举人
I've been experimenting with OCR, an iPod touch and lenses. Objective was being able to capture text, up to 4 characters at a time (chengyu), over as wide a range of character sizes as possible.

There is a lot of leeway in how much magnification to use. More magnification shifts OCR use towards smaller point sizes. However, at higher magnification you will need a (very) steady hand to focus (smaller depth of field). As an experiment, you might try holding a small magnifier, eye loupe or linen tester in front of the ipod touch camera.

In general, the best lens is the lens with the lowest magnification which still does the job. Good results were obtained with lenses with a magnification of x3 to x4. A magnification of x3 to x4 corresponds to a diopter of +12 to +16, or a focal distance of 8.3 to 6.25 cm. In the end, I settled upon a lens with a focal distance of 7.5 cm (magnification x3.33). (Achromat f = 75mm) This allows OCR on characters between 6 pt and 24 pt, with the "sweet spot" being about 12 pt. To get meaningful results for 6pt text, I need to select the "camera icon" and zoom in; for 24pt text I need to select the "camera icon" and zoom out.

To focus the image, adjust the distance between iPod and book. You'll get a sharp image when the distance is the focal distance. Once focused, it helps if the iPod rests on something: if you hold the iPod in your hand the image can be jittery, especially at small point sizes. But in the end, there is only so much one can do. iPod Touch camera performance is profoundly unexciting.

Summarizing: if you have an iPod OCR can be made to work by adding a lens, but the user experience is not as intuitive as with an iPhone.
 

Attachments

  • ipod_touch.jpg
    ipod_touch.jpg
    130.2 KB · Views: 1,832

character

状元
^ Good results and information.

I've not tried it for OCR yet, but the Belkin Video Stand+Charge Sync seems like it will be good for holding the iDevice steady and letting one move material to be OCR'd.
 

Dr.Grace

秀才
I don't have one of those Belkin stands, but I suppose you could clamp it to a gooseneck microphone stand to hold it in a horizontal position and place the material for OCR flat on the desk.
 
Top