Finding position of block is so slow

[azs-rahimi]

Hi
I need to find position of each text block in image. I've read issue #64 and #81 and according to those, my code is:

Tesseract.Pix pix = Tesseract.PixConverter.ToPix(testimg);
using (var page = engine.Process(pix))
{
using (var iter = page.GetIterator())
{
do
{
if (iter.TryGetBoundingBox(Tesseract.PageIteratorLevel.Block, out blockBounds))
{
imgblock.Draw(new Rectangle(blockBounds.X1, blockBounds.Y1, blockBounds.Width, blockBounds.Height), new Bgr(255, 0, 0), thickness);
}
} while (iter.Next(Tesseract.PageIteratorLevel.Para));
}
}

It takes about 20 seconds for each image. Their size are 1868*2416. Is there another way to detect block size instead of "iter = page.GetIterator()"? This command is very slow for me.

Regards
Zahra

[charlesw]

You can't, the page iterator is how you extract this information. There is
no alternative.

Is it the getiterator call itself that's taking 20 seconds or the iteration
loop?
On 25 Nov 2014 16:26, "azs-rahimi" notifications@github.com wrote:

Hi
I need to find position of each text block in image. I've read issue #64
#64 and #81
#81 and according to those,
my code is:

Tesseract.Pix pix = Tesseract.PixConverter.ToPix(testimg);
using (var page = engine.Process(pix))
{
using (var iter = page.GetIterator())
{
do
{
if (iter.TryGetBoundingBox(Tesseract.PageIteratorLevel.Block, out
blockBounds))
{
imgblock.Draw(new Rectangle(blockBounds.X1, blockBounds.Y1,
blockBounds.Width, blockBounds.Height), new Bgr(255, 0, 0), thickness);
}
} while (iter.Next(Tesseract.PageIteratorLevel.Para));
}
}

It takes about 20 seconds for each image. Their size are 1868*2416. Is
there another way to detect block size instead of "iter =
page.GetIterator()"? This command is very slow for me.

—
Reply to this email directly or view it on GitHub
#138.

[azs-rahimi]

Calling the getitetator is about 20 sec for me. The loop is not slow (Up to one second).
Thank you again

[charlesw]

I think I know why it's taking that long and that's because getiterator is
actually doing the processing (on demand). You can verify that by just
calling Page.GetText() it should take around the same time.

If this is the case then you might be able to speed it up by changing some
parameters and try to trade accuracy for performance eg the page seg mode
or EngineMode.
On 25 Nov 2014 18:53, "azs-rahimi" notifications@github.com wrote:

Calling the getitetator is about 20 sec for me. The loop is not slow (Up
to one second).
Thank you again

—
Reply to this email directly or view it on GitHub
#138 (comment).

[azs-rahimi]

I set EngineMode to "Default". It causes speeding up the getitetator calling .
Thank you Charles for your help

[azs-rahimi]

I need more speed. Its run time is about 5 seconds. No other suggestion?

[charlesw]

Are you doing the processing concurrently? For instance multiple documents
at once.
On 27 Nov 2014 02:22, "azs-rahimi" notifications@github.com wrote:

I need more speed. Its run time is about 5 seconds. No other suggestion?

—
Reply to this email directly or view it on GitHub
#138 (comment).

[azs-rahimi]

No, this time is just for one getiterator calling, and there is no other
process while it runs.

On Wed, Nov 26, 2014 at 11:03 PM, Charles Weld notifications@github.com
wrote:

Are you doing the processing concurrently? For instance multiple documents
at once.
On 27 Nov 2014 02:22, "azs-rahimi" notifications@github.com wrote:

I need more speed. Its run time is about 5 seconds. No other suggestion?

Reply to this email directly or view it on GitHub
#138 (comment).

Reply to this email directly or view it on GitHub
#138 (comment).

Azam Zahra Rahimi
M.Sc. Student at Ferdowsi University of Mashhad
Computer Engineering - Artificial Intelligence

[charlesw]

OK, this might just be how long it takes to process the image. Might be worth trying just running it through the tesseract.exe and seeing how long that takes.

[azs-rahimi]

Do you mean I install the tesseract binary and run with it?

[azs-rahimi]

I install tesseract -3.02.02 and run it. "tesseract MyPage.png outPage.txt -l eng"
It takes lower times about 3 seconds. But I can't find position of blocks. When I change the "psm" option, my outPage is 0 byte.

[charlesw]

Yes that's what I ment, idea was to get a baseline timing. I also forgot
that you could also use the gethocr functionality which should return an
xml blob with these values. Might be marginally faster given there
wouldn't be as much interop involved. However you would then need to parse
the output.

What's the use case here? I would have thought a couple of seconds would
have been acceptable in most cases?
On 29 Nov 2014 15:59, "azs-rahimi" notifications@github.com wrote:

I install tesseract -3.02.02 and run it. "tesseract MyPage.png outPage.txt
-l eng"
It takes lower times about 3 seconds. But I can't find position of blocks.
When I change the "psm" option, my outPage is 0 byte.

—
Reply to this email directly or view it on GitHub
#138 (comment).

[azs-rahimi]

Could you please tell me how can I use the gethocr function.
I want to detect blocks and present to user. Then user selects any of them
(a desired block) for OCR process.

On Sat, Nov 29, 2014 at 1:02 PM, Charles Weld notifications@github.com
wrote:

Yes that's what I ment, idea was to get a baseline timing. I also forgot
that you could also use the gethocr functionality which should return an
xml blob with these values. Might be marginally faster given there
wouldn't be as much interop involved. However you would then need to parse
the output.

What's the use case here? I would have thought a couple of seconds would
have been acceptable in most cases?
On 29 Nov 2014 15:59, "azs-rahimi" notifications@github.com wrote:

I install tesseract -3.02.02 and run it. "tesseract MyPage.png
outPage.txt
-l eng"
It takes lower times about 3 seconds. But I can't find position of
blocks.
When I change the "psm" option, my outPage is 0 byte.

Reply to this email directly or view it on GitHub
#138 (comment).

Reply to this email directly or view it on GitHub
#138 (comment).

Azam Zahra Rahimi
M.Sc. Student at Ferdowsi University of Mashhad
Computer Engineering - Artificial Intelligence

[charlesw]

Okay, in this case I'd just do the following:

1. Call Page.AnalyseLayout and use the resulting iterator to render your
   regions
2. Once user selects the region they want set the region of interest and
   call get text or getiterator if you need additional information.

I don't think you'll be able to speed it up more than that.
On 30 Nov 2014 05:11, "azs-rahimi" notifications@github.com wrote:

Could you please tell me how can I use the gethocr function.
I want to detect blocks and present to user. Then user selects any of them
(a desired block) for OCR process.

On Sat, Nov 29, 2014 at 1:02 PM, Charles Weld notifications@github.com
wrote:

Yes that's what I ment, idea was to get a baseline timing. I also forgot
that you could also use the gethocr functionality which should return an
xml blob with these values. Might be marginally faster given there
wouldn't be as much interop involved. However you would then need to
parse
the output.

What's the use case here? I would have thought a couple of seconds would
have been acceptable in most cases?
On 29 Nov 2014 15:59, "azs-rahimi" notifications@github.com wrote:

I install tesseract -3.02.02 and run it. "tesseract MyPage.png
outPage.txt
-l eng"
It takes lower times about 3 seconds. But I can't find position of
blocks.
When I change the "psm" option, my outPage is 0 byte.

Reply to this email directly or view it on GitHub
<
https://github.com/charlesw/tesseract/issues/138#issuecomment-64941746>.

Reply to this email directly or view it on GitHub
<#138 (comment)
.

Azam Zahra Rahimi
M.Sc. Student at Ferdowsi University of Mashhad
Computer Engineering - Artificial Intelligence

—
Reply to this email directly or view it on GitHub
#138 (comment).

[azs-rahimi]

I've used Page.AnalyseLayout() instead of Page.GetIterator(). It seems this function is faster than the GetIterator.
Thank you again Charles for your help

[charlesw]

No problem
On 30 Nov 2014 17:21, "azs-rahimi" notifications@github.com wrote:

I've used Page.AnalyseLayout() instead of Page.GetIterator(). It seems
this function is faster than the GetIterator.
Thank you again Charles for your help

—
Reply to this email directly or view it on GitHub
#138 (comment).