OCR!

mikelove

皇帝
Staff member
character said:
"We got a chance to spend a little time with Contour's upcoming Live Viewfinder for the ContourGPS, an app that finally lets us know for sure which way our little cameras are pointing. The app makes a Bluetooth connection to the camera and pulls down footage in real-time [...]"

http://contour.com/camera/contour_gps

Not clear if they will open up the API.

Also not clear how well it does macro focus - one of the demo videos on their website seems promising but it's still pretty far away from its target. Given that it's mainly being marketed to outdoor sports enthusiasts interested in recording their various exploits, I'm not sure if they'd want to put much effort into letting it focus on close-up text...
 

character

状元
Yeah, it might not be the best camera for OCR, but the more BT cameras the better -- one of them will work well enough.
 

tanyahart

秀才
Owle Bubo
http://www.youtube.com/watch?v=Qdi2UkvWyOs
This is VERY funky looking iPhone add-on. - A wide angle lens for iPhone film making with tripod/monopod screw-holes. It's a bit of overkill for OCR capture maybe. But it still looks like a beautiful thing. You could set it up with a little desktop tripod and you'd be off!

Tanya.
 

Mingdemen

Member
Re: 'Hanged'

Hi. I annoying trouble I have is that on many occasions, the OCR seem to 'hang'. It does not move. In fact, in other instances, when I am holding it in the 'portrait' posn, it shows the ''landscape' posn and vice-versa. Hope I make sense. Is there a way to rectify this?

Thank you.
 

mikelove

皇帝
Staff member
That one's news to us... what type of device are you using it on? Does the video feed continue but the OCR stop updating or does the video hang as well?
 

Mingdemen

Member
Hi. I am using Iphone 4. If by video feed you mean whether what I see on the screem, well, yes, it appears ok. The only thing that 'hangs' is that it keeps show yi yi for whichever word it is supposed to dicipher.

Thanks,
 

mikelove

皇帝
Staff member
Mingdemen said:
Hi. I am using Iphone 4. If by video feed you mean whether what I see on the screem, well, yes, it appears ok. The only thing that 'hangs' is that it keeps show yi yi for whichever word it is supposed to dicipher.

Does the 一一 keep moving or does it remain still / frozen after it's recognized? Try this: go into the Settings / OCR / Lookup Words screen and turn "Sauvola binarization" to whatever option it isn't on now - does that help any?
 

keyclick

秀才
Mike,

I have an Ipod touch 3rd gen (latest model before the recent retina display/crapcamera upgrade model) and wanted to try the OCR in 'saved image mode'. I just downloaded the 2.2.2 update of Pleco, and I thought I read in the App Store Updates page for Pleco (as I was reading the revision notes and pressing "download") that there was now a usable demo for OCR available in non-camera devices, and that the improvements to the stored-image version of OCR were such that it was now a worthwhile feature even without a camera.

I went to the "I want to override the appstore to allow me to download OCR to a non Iphone4 device" page, and registered my UUID, but the OCR item is still greyed out in the inapp purchase section.

Reading the requirements there, it says ipod 3rd gen is OK, but also says it needs a camera and multitasking. Is the newest ipod touch considered 4th gen?

Anyway, any help clearing up this confusion? Also, I did not see it in a cursory search, but is there a revision notes page on your website? After downloading an updata (2.2.2) the revision notes that are listed in the Updates page of App Store disappear as soon as you update, and the main pleco entry on the appstore only has the introductory description, not the x.x revision information. In this case, I wish I could re-read what I thought I read in the rev notes to see if I was wrong :)

Thanks again! Cant wait to try out the lyrics reader... was hoping for such a thing. Now just need to upload some of my lyric-included podcasts!

Greg
 

keyclick

秀才
thanks, character! My ipod is model MC086LL which has been billed as 3rd gen... but the info you provided sort of a reminder...
I think the 8gb model was left in the line up when 3rd gen came out with 16 and higher gig models and the increased RAM. The 8 gig was rebranded as 3rd gen to be part of the 3rd gen family but in name only, not in horsepower.

I remember when I upgraded to ios4.x and was saddened to see no desktop wallpaper or true multitasking that I had hoped for.
So I guess its not a true 'hardware 3rd gen'.

So, mike, is the RAM now the limiting factor (besides the camera) that makes this model of ipod touch not OCR capable?

Thanks, guys!
 

character

状元
I can't remember for certain how Apple handled the difference (it was probably clear enough to prevent lawsuits, at least), but perhaps Mike should add some language to make it clear the only 3rd generations iPod Touches to support OCR are the 16/32/64GB ones.
 

neobarney

Member
I'm a very big fan of Pleco, and have just downloaded the OCR feature, which looks great and very accurate compare to some other OCR softwares.
So I was wondering if you guys had any plan to add a new feature, allowing users to send for example a PDF document into Pleco, containing Chinese text in the image format, and then use the OCR feature to tranform the content to the text format ?

I guess the checking process could be kind of similar to the flashcards import : if some character is not clearly recognized, then the user could be prompted to select the entry manually by itself, and decide or not to apply the same action to any future occurrence of the same character in the same text.

Maybe it would provide a feature much more powerful than other OCR, because it is adding the ability to first segment the text in order to strengthen the fiability of the recognition process?

For me, I would really appreciate such a feature, cause I like to use DVD movies to study Chinese. Sometime, it is impossible to find the subtitles in the text format over the web, and then the only solution is to extract the subtitles bitmaps, then paste them into a word file, save it as a pdf and finally use Abby finereader to transform the content to the text format.
The main issue is that Abby fine reader looks far weaker than Pleco OCR, I guess mainly because it doesn't have the dictionaries included inside Pleco.

Many thanks for your great work on Pleco!!

Mickael
 

mikelove

皇帝
Staff member
keyclick said:
So, mike, is the RAM now the limiting factor (besides the camera) that makes this model of ipod touch not OCR capable?

Yes, unfortunately - ironically, 128 on an iPod would be just barely enough for live OCR (it's not enough on the iPhone 3G because of the extra RAM occupied by the always-on phone libraries) but 256 is a bare minimum for still image; even with compression, in order to make performance decent we need roughly 3x the number of pixels in the image worth of bytes just for image storage / image buffers, along with the copious amounts of memory used by the actual recognizer.

character said:
I can't remember for certain how Apple handled the difference (it was probably clear enough to prevent lawsuits, at least), but perhaps Mike should add some language to make it clear the only 3rd generations iPod Touches to support OCR are the 16/32/64GB ones.

Good idea, yes.

neobarney said:
So I was wondering if you guys had any plan to add a new feature, allowing users to send for example a PDF document into Pleco, containing Chinese text in the image format, and then use the OCR feature to tranform the content to the text format ?

It's on our to-do list, yes - it's fairly easy to render PDFs in any resolution we want, it's just trickier than opening images in regular JPEGs and such.

neobarney said:
I guess the checking process could be kind of similar to the flashcards import : if some character is not clearly recognized, then the user could be prompted to select the entry manually by itself, and decide or not to apply the same action to any future occurrence of the same character in the same text.

It's actually tough to know when a character is "clearly recognized" - the nature of the algorithm is such that the average statistical accuracy for a correct match varies depending on the character / font / resolution / etc to a point where we can't really come up with a clear consistent way to know which characters we ought to flag. (we did a lot of testing with this before abandoning it)

neobarney said:
Maybe it would provide a feature much more powerful than other OCR, because it is adding the ability to first segment the text in order to strengthen the fiability of the recognition process?

It could certainly help, though we're a bit limited in terms of what sorts of hints we can plug into the algorithm (core stuff is optimized like crazy and very tough to change).

neobarney said:
For me, I would really appreciate such a feature, cause I like to use DVD movies to study Chinese. Sometime, it is impossible to find the subtitles in the text format over the web, and then the only solution is to extract the subtitles bitmaps, then paste them into a word file, save it as a pdf and finally use Abby finereader to transform the content to the text format.

What do those bitmaps generally look like? Are they just the characters in a plain color against a plain background or are there lots of borders / shadows / etc? That affects our recognizer's ability to work accurately...
 
OCR seems to fail for "not dark" characters. I get good results for even very stilistic signs, but fails me completely on white letters. Not important, just my notice.
 

Attachments

  • Unnamed.jpg
    Unnamed.jpg
    8.6 KB · Views: 2,490

mikelove

皇帝
Staff member
sofamensch said:
OCR seems to fail for "not dark" characters. I get good results for even very stilistic signs, but fails me completely on white letters. Not important, just my notice.

Does it help if you turn on the Sauvola Binarization option in Settings / OCR / Live Video / Mode-specific / Lookup words?
 

dustpuppy

榜眼
To the pleco team: you completely nailed the still image OCR features. It's amazing that after a couple of months, I get a completely new feature set. I'm so happy with pleco, it's the best application on my iphone period. Thank you, thank you !!
 

dustpuppy

榜眼
Note: I now prefer to use the OCR feature using still images than with the live OCR. It's just much more stable, the camera doesn't go out of focus, etc.
 

dustpuppy

榜眼
I'm studying from my chinese textbook with OCr still image (scroll lookup words), I can't believe how ****ing awesome it is, how effortlessly I can read and lookup characters when I don't know them. Congratulations !! And sorry about posting three times, I'm a software engineer and when I see software with this outstanding level of quality, I just have a lot of praise to give.
 

mikelove

皇帝
Staff member
dustpuppy said:
I'm studying from my chinese textbook with OCr still image (scroll lookup words), I can't believe how ****ing awesome it is, how effortlessly I can read and lookup characters when I don't know them. Congratulations !!

Thanks!
 
Seems the iPad 2 has no auto-focus cam. I wonder how that will work with Pleco then. It works somehow on my iPod, but it's way better on an iPhone (with auto-focus - and the light also helps a lot).
 
Top