Comments on: Mobile OCR input: “Fully automatic” and reality

By: folha

folha — Mon, 23 Aug 2010 01:48:08 +0000

free ocr is a online ocr service. You can have a try.

By: jk

jk — Wed, 09 Jun 2010 16:23:31 +0000

get forced closed all the time doesnt work for me on htc desire

By: luke

luke — Mon, 22 Mar 2010 11:51:17 +0000

I’ve posted GOCR compiled for android to your http://code.google.com/p/wordsnap-ocr/issues/detail?id=3

By: Sam Joseph

Sam Joseph — Sat, 20 Mar 2010 01:51:33 +0000

Great app! I’ve been trying to compile and deploy the app to my android phone from the source code, but I run into this problem:

03-19 00:32:56.057: ERROR/AndroidRuntime(3249): Uncaught handler: thread
OCRThread exiting due to uncaught exception
03-19 00:32:56.067: ERROR/AndroidRuntime(3249):
java.lang.UnsatisfiedLinkError: nativeMin
03-19 00:32:56.067: ERROR/AndroidRuntime(3249): at
net.bitquill.ocr.image.GrayImage.nativeMin(Native Method)

Issue further documented in google code

http://code.google.com/p/wordsnap-ocr/issues/detail?id=2

Any help greatly appreciated!

By: Alex

Alex — Sat, 23 Jan 2010 14:47:55 +0000

P.S. Given that Dalvik is only interpreting Java, I no longer wonder why Android is only as fast as J2ME on much less powerful devices… or that the same code ported via xmlvm to iPhone is about 20x faster!

By: Alex

Alex — Sat, 23 Jan 2010 14:44:23 +0000

I’ve tested WordSnap on the Samsung i7500 w/ Android 1.5… works quite well! The automated mode is not working, it switches the LED light on and off (and when it is on, the image is much too bright). The manual model works as expected, although it seems the system is biased towards dark letters on a bright background and works not at all for bright letters on a dark background.

I think that porting GoCR is a waste of time. The only thing fast enough for such small platforms would be a convolutional neural network trained to detect single letters (possibly including rotated letters). This simplifies to a large set of convolutions of the image,which can be done very efficiently in phase space, works without a large dictionary, and would be fast enough to scan paragraphs of text in real-time. However, such systems are very hard to train and the availability of sufficient training data is a real challenge…

By: Hmm

Hmm — Tue, 03 Nov 2009 19:27:57 +0000

“Binarization” seems like a bad idea to me. You’re throwing away information that you could use to read the letters more accurately. Is this normally done for OCR?

By: Jamie

Jamie — Thu, 24 Sep 2009 04:08:51 +0000

Just came across your site accidentally looking for a mobile OCR. Felt the need to comment. Really great work! Would love to see this working realtime.

By: Neil

Neil — Wed, 23 Sep 2009 13:08:17 +0000

On your TODO list is:
Look into porting one of the lighter-weight engines (GOCR, perhaps via Conjecture) to Android

The one you didn’t mention, Tesseract, has been ported to Android, or at least a JNI wrapper has been written:
http://www.itwizard.ro/mezzofanti-augmented-reality-through-text-recognition-146.html

By: Charalampos Tsourakakis

Charalampos Tsourakakis — Mon, 07 Sep 2009 05:37:06 +0000

Really nice work Spiro!