«

Hi group!

I need to extract some text (well numbers actually) from a bunch of
similarly looking .png images. After extraction the numbers will be fed to a
Python script for further processing. Any good ideas on how to go about with
this? I have no idea whatsoever about how to extract the numbers out of the
images...

Thanks in advance,

Henrik

"Henrik Berg Nielsen" <> writes:

> I need to extract some text (well numbers actually) from a bunch of
> similarly looking .png images. After extraction the numbers will be fed to a
> Python script for further processing. Any good ideas on how to go about with
> this? I have no idea whatsoever about how to extract the numbers out of the
> images...

OCR is the TLA you're looking for ("Optical Character Recognition").

Dunno if there are any good free OCR engines. With these sorts of
hard algorithms, you tend to get what you pay for.


John

Henrik Berg Nielsen <> spake thusly:
>
> I need to extract some text (well numbers actually) from a bunch of
> similarly looking .png images. After extraction the numbers will be fed
> to a Python script for further processing. Any good ideas on how to go
> about with this? I have no idea whatsoever about how to extract the
> numbers out of the images...
>
This might help you out...
http://www.pricelessware.org/2003/PL...tm#Convert-OCR

I'm not sure if it does PNG, you might have to convert the file to tiff or
bmp or something.


--
Audio Bible Online:
http://www.audio-bible.com/

In article <wbDeb.2223$>, Henrik Berg Nielsen wrote:
> Hi group!
>
> I need to extract some text (well numbers actually) from a bunch of
> similarly looking .png images. After extraction the numbers will be fed to a
> Python script for further processing. Any good ideas on how to go about with
> this? I have no idea whatsoever about how to extract the numbers out of the
> images...
>


http://www.claraocr.org/

John> OCR is the TLA you're looking for ("Optical Character Recognition").

John> Dunno if there are any good free OCR engines. With these sorts of
John> hard algorithms, you tend to get what you pay for.

Which often means there's a piece of free software out there which works
better than the most expensive commercial solutions. <wink>

A little googling suggests this might be a candidate:

http://www.claraocr.org/

I have no idea if there's an exported library and/or a Python wrapper, but
it's probably worth a look.

Skip

"Henrik Berg Nielsen" <> wrote:
>
>I need to extract some text (well numbers actually) from a bunch of
>similarly looking .png images. After extraction the numbers will be fed to a
>Python script for further processing. Any good ideas on how to go about with
>this? I have no idea whatsoever about how to extract the numbers out of the
>images...

Are you hoping to extract the "password" characters from the pictures
presented by the whois checks? If so, you should give up now, because
those images are SPECIFICALLY designed to make them almost impervious to
automated recognition.
--
- Tim Roberts,
Providenza & Boekelheide, Inc.

Henrik Berg Nielsen wrote:
> Hi group!
>
> I need to extract some text (well numbers actually) from a bunch of
> similarly looking .png images. After extraction the numbers will be fed to a
> Python script for further processing. Any good ideas on how to go about with
> this? I have no idea whatsoever about how to extract the numbers out of the
> images...

Hi,
I'm dealing with similar problem now. My pictures are very complicated
(construction drawings). I am trying to use gamera
(http://dkc.jhu.edu/gamera/) for OCR and it seems very promising.

--
-- Lukas

On Wed, 01 Oct 2003 20:25:45 -0700, Tim Roberts <> wrote:

>"Henrik Berg Nielsen" <> wrote:
>>
>>I need to extract some text (well numbers actually) from a bunch of
>>similarly looking .png images. After extraction the numbers will be fed to a
>>Python script for further processing. Any good ideas on how to go about with
>>this? I have no idea whatsoever about how to extract the numbers out of the
>>images...
>
>Are you hoping to extract the "password" characters from the pictures
>presented by the whois checks? If so, you should give up now, because
>those images are SPECIFICALLY designed to make them almost impervious to
>automated recognition.
Sounds interesting as a problem, but I wouldn't want to create a skeleton key
for any bad guys

Regards,
Bengt Richter