Detecting simple phone number

[Flipz]

Hi,

I installed tesseract to do some simple OCR on a very basic image - a
phone number in Arial 11pt in a png format (http://www.autohop.bg/OCR/
phone.png) which I convert to a uncompressed tiff using:

convert -monochrome -normalize ./phone.png ./phone.tif

When I run it through tesseract I get an empty file from the following
command:

/usr/bin/tesseract ./phone.tif phone

I tried training the program but I cant get the whole logic around the
box files. I only need to detect the numbers in the png. Do I need to
additionally train tesseract for that or it is built in the current
eng support? I looked very simple from the tesseract documentation but
apparently I am missing out on something because I get an empty
phone.txt file.

Any help will be appreciated.

Thanks,

CJ

[Yury Tarasievich]

For tesseract to produce something, you'll need
an image with a resolution on the order of 100-s
of dpi. I find that 300 dpi works
satisfactorily, but 200 dpi seems also okay. And
you have something like 72 dpi image there.

Flipz wrote:
> Hi,
>
> I installed tesseract to do some simple OCR on a very basic image - a
> phone number in Arial 11pt in a png format (http://www.autohop.bg/OCR/
> phone.png) which I convert to a uncompressed tiff using:

...

[Flipz]

Do you know a way to upscale this png to 300dpi through a Linux
command (maybe in convert?). As far as I know web images are mostly
72dpi and this means I won't be able to use tesseract for any of them,
correct?

If this image is at 300dpi, do you think it is managable for tesseract
to extract the numbers?

On Jul 4, 1:10 pm, Yury Tarasievich <yury.tarasiev...@gmail.com>
wrote:

[Yury Tarasievich]

Flipz wrote:
> Do you know a way to upscale this png to 300dpi through a Linux
> command (maybe in convert?). As far as I know web images are mostly
> 72dpi and this means I won't be able to use tesseract for any of them,
> correct?
>
> If this image is at 300dpi, do you think it is managable for tesseract
> to extract the numbers?

I don't really know, I haven't yet try to do the
full OCR cycle with an 4x scaled image. I speak
only from my limited experience with scans and
digital photos processing in tesseract. E.g., I
had some 150 dpi photos of book pages, and I
scaled them 2x in Gimp with standard settings,
and those tesseract processed fairly well.

[WHEAT]

hi,

We're Sky Studio, profressional OCR develop team, we have check the

http://www.autohop.bg/OCR/phone.png , no problem at all that we can

make a OCR　for this simple phone number pictures. with >=95% success

rate, within 1 days ! We will only charge 100usd for this OCR development, and with 1 year free tech support. you can choose DLL or

 commandline program. we will provide detail samples on how to use the dll or commandline program , in c#, vb, delphi .

 

Looking forward to your reply .

 

Best Regards

Richard

Sky Studio Inc.

http://www.yahoobot.net/

7/5/2009

---

200万种商品,最低价格,疯狂诱惑你

[Yury Tarasievich]

Flipz wrote:
> Do you know a way to upscale this png to 300dpi through a Linux
> command (maybe in convert?). As far as I know web images are mostly
> 72dpi and this means I won't be able to use tesseract for any of them,
> correct?
>
> If this image is at 300dpi, do you think it is managable for tesseract
> to extract the numbers?

Can't see why not. And upscaling may indeed be
done with convert or, possibly, Gimp. I can't
point out the best algorithm for the upscaling,
however (you know, biliniear, bicubic etc.)

--

[Flipz]

I did it with convert, it has a very nice set of upscaling algorithms
and I managed to pull it off to 300dpi which gives 98%+ accuracy
rate..

Thanks for all your help, especially Yury's for pointing me to the
direction of upscaling the image, I was trying to create a custom
tesseract box for something already built in the eng distribution of
the software :)

[zhi]

Hi Flipz,

I run into the same problem as yours. Can you pleas tell me how to
increase the dpi of an image in linux with convert and how to check
the dpi of an image with command?

Thanks

[Flipz]

Here are the commands I use:

$phone_image = "test.png";
$phone_image_1 = "test_upscale.png";
$phone_image_2 = "test_upscale.tif";
$phone_extract = "test_phone";

exec("/usr/bin/convert -filter Lanczos -resample 300x300 -enhance ".
$phone_image." ".$phone_image_1);
exec("/usr/bin/convert -monochrome -normalize ".$phone_image_1." ".
$phone_image_2);
exec("/usr/bin/tesseract ".$phone_image_2." ".$phone_extract);

> > > --- Скриване на цитирания текст -
>
> - Показване на цитирания текст -

[zhi]

Flipz, thanks so much. and Is there anyway to check the dpi of an
image with command?