Re: If I pass a binary (black-white) image into tesseract-ocr, will tesseract process and change the image on its own way?

[Patrick Questembert]

I think you are right - ScanBizCards is passing a black & white image to Tesseract and we are pretty sure Tesseract doesn't change the image (empirically - we never dug in to make sure).

Patrick

On Sun, Dec 16, 2012 at 9:44 PM, Linda Li <codingpo...@gmail.com> wrote:

If I pass a binary (black-white) image into tesseract-ocr, will tesseract process the image on its own way?

It seems tesseract ocr uses Otsu global thresholding method (not looking into the code carefull yet, but saw the key word “otsu” in the source code)

If so, the Otsu method will not change the binary (black-white) image.

So in my understanding, tesseract-ocr will not change the input binary image.

Just ask to make sure...

Thanks.

--
You received this message because you are subscribed to the Google
Groups "tesseract-ocr" group.
To post to this group, send email to tesser...@googlegroups.com
To unsubscribe from this group, send email to
tesseract-oc...@googlegroups.com
For more options, visit this group at
http://groups.google.com/group/tesseract-ocr?hl=en

--
Patrick Questembert, ScanBizCards

+1-917-250-4177 | www.scanbizcards.com

twitter.com/ScanBizCards | www.facebook.com/ScanBizCards

Just released: Power Contacts - http://itunes.apple.com/us/app/power-contacts/id476986356?mt=8

[Patrick Questembert]

Yes, we convert the image to black and white, that's the end result of any image processing for the purpose of OCR - Tesseract recognition has to work on a black & white image so either your app creates the b&w image or you let Tesseract create it. We had to create the b&w image because we found Tesseract's image processing to be inadequate for most business card images. Hard to fully describe all the details of our image processing but it include stages like detecting areas where patterns are present (as opposed to background), local adaptive thresholding and non-text elimination (the later is something we don't do too well yet - but Tesseract is usually not too confused by non-text patterns in the b&w image). You can see the b&w image we generate by turning ON a setting of ScanBizCards called "Show Processed Image" (available only in our iOS version, not the Android version).

Note also that in the results you see in ScanBizCards it's not just about the b&w image: we also have 70,000 lines of code examining the text results returned by Tesseract and comparing to our home-brewed OCR tests then picking one or the other for each letter.

Patrick

On Sat, Jan 12, 2013 at 9:02 AM, Tin Siladin <tsil...@gmail.com> wrote:

Patrick,

I ran into your answer and I think you might be able to help me given that your scanbizcards app is doing similar processing to what I need (don't worry, no competition :) ).

I'm using tesseract-ocr on android, taking pictures of portions of receipts in grayscale mode (EFFECT_MONO, if available on device), and converting the bitmap to ARGB_888 for tesseract processing.  I'm getting pretty good results, but am trying to improve accuracy.

Do you know whether you do any other preprocessing of the bitmap before passing to tesseract, like perhaps converting to black/white (you mentioned in your above answer you're passing "black & white image").  I tried out your app and it seems to give better results than mine on similar types of images.  Very nice app, BTW.

Thanks,

Tin

[Sven Pedersen]

Many people get good results using the ImageMagick library/command line tool for binarization and other pre-processing. It's free and you can use the library for commercial products (Apache 2.0 license like Tesseract).

--Sven

On Sat, Jan 12, 2013 at 9:16 AM, Tin Siladin <tsil...@gmail.com> wrote:

Thanks for your response.  In the meantime, I read up on many of your answers in this group which will be very valuable to me. Thank you for sharing your expertise. 

I have in my incoming bitmap a single line of text (mostly numbers), so I don't need to worry about  most of the details you do. I believe the variances in accuracy occur due to different lighting/brightness conditions under which the picture is taken so sometimes the incoming grayscale image is overall darker, etc.  I've played around with contrast settings of the camera, and that helps as well, but am still trying to improve. Preprocessing to black and white was just one idea I had in mind. Needless to say, I'm fairly new to tesseract, so read up on possibilities of setting configuration parameters/variables which may help me. 

If you guys have nothing better to do, the "Show Processed Image" option would be nice to have in Android as well.  :)

Thanks again,

Tin

--
``All that is gold does not glitter,
  not all those who wander are lost;
the old that is strong does not wither,
  deep roots are not reached by the frost.
From the ashes a fire shall be woken,
  a light from the shadows shall spring;
renewed shall be blade that was broken,
  the crownless again shall be king.”