beam search failed?!

[zhi]

Hi guys,

After no lucking trying to install Ocropus on my Centos box, I now
followed closely and successfully installed it on my ubunbu 9.04. I
tested about 10 images and none of them seems to work. I copied some
of output and hope someone can help me out.

I got

test@ubuntu:~/ocrtest$ ocropus page 32047.png
Ag3A
For a quota0on
0oca'I: 1890 6i7 6i7
Exis'ng Cutome's
tocan: 1890 618 618
[beam search failed]
[beam search failed]
[beam search failed]
[beam search failed]
[beam search failed]


test@ubuntu:~/ocrtest$ ocropus page LimeDaleybusinesscard.png
Jon Daley
131 Denny Road
Uaencia, PA 1 6059
jon4imedaleycom
412-360m844
'yyrme wv imedaleycom
[info] WARNING: feature extraction failed [28]: CHECK ocr-line/
linerec.cc:350 b.height()<pgetf("maxheight")
[beam search failed]
. Embedded Sob8re
. System Design
. Testing Automation
. FTP and Web Hosting
. Personal Technical Support
. High'y Customized Solutions
. Industrial Controls . Individuals to Medium-Sized Businesses
terminate called after throwing an instance of 'glinerec::BadTextLine'
Aborted


test@ubuntu:~/ocrtest$ ocropus page LimeDaleybusinesscard.png
Jon Daley
131 Denny Road
Uaencia, PA 1 6059
jon4imedaleycom
412-360m844
'yyrme wv imedaleycom
[info] WARNING: feature extraction failed [28]: CHECK ocr-line/
linerec.cc:350 b.height()<pgetf("maxheight")
[beam search failed]
. Embedded Sob8re
. System Design
. Testing Automation
. FTP and Web Hosting
. Personal Technical Support
. High'y Customized Solutions
. Industrial Controls . Individuals to Medium-Sized Businesses
terminate called after throwing an instance of 'glinerec::BadTextLine'
Aborted

[Thomas Breuel]

Can you submit an issue report on ocropus.org and attach the image, please?

Based on that, we can determine whether it's an actual problem, or
simply a document type that OCRopus isn't trained to recognize yet.
We can then prioritize it accordingly.

"Beam search failed" simply means one of two things. Either something
went wrong with the segmentation of the input line, or there was no
reasonable way for OCRopus to recognize it using its current models.

Tom

[zhi]

submitted, hope it can help in someway.

Thanks

On Jul 1, 2:40 pm, Thomas Breuel <tmb...@gmail.com> wrote:
> Can you submit an issue report on ocropus.org and attach the image, please?
>
> Based on that, we can determine whether it's an actual problem, or
> simply a document type that OCRopus isn't trained to recognize yet.
> We can then prioritize it accordingly.
>
> "Beam search failed" simply means one of two things.  Either something
> went wrong with the segmentation of the input line, or there was no
> reasonable way for OCRopus to recognize it using its current models.
>
> Tom
>

[Thomas Breuel]

Thanks. There's nothing wrong in principle with those inputs (the
text is obviously nice and clean). However, this kind of input is not
what OCRopus is currently trained or tested on (neither is Tesseract,
which also gives fairly noisy output).

OCRopus should work better on them in 0.5, after it has been trained
on more kinds of inputs.

Tom

[zhi]

thanks Tom, looking forward to trying out 0.5

On Jul 1, 3:20 pm, Thomas Breuel <tmb...@gmail.com> wrote:
> Thanks.   There's nothing wrong in principle with those inputs (the
> text is obviously nice and clean).  However, this kind of input is not
> what OCRopus is currently trained or tested on (neither is Tesseract,
> which also gives fairly noisy output).
>
> OCRopus should work better on them in 0.5, after it has been trained
> on more kinds of inputs.
>
> Tom
>