tesseract: broken pipe ⇒ Falling back on Paperwork's heuristic

[tYYGH]

I had once again orientation detection failing, and so I decided to ditch the Archlinux package, and install paperwork/stable from git as explained here:
https://github.com/jflesch/paperwork/blob/stable/doc/install.devel.markdown#paperwork-in-a-python-virtualenv
Unfortunately, this did not solve the issue. Here are the relevant logs:

INFO   paperwork.frontend.mainwindow.scan Scan started
INFO   paperwork.frontend.util.canvas.drawers Drawer: Target area: (105, 0) ((527, 727)) << (105, 0) ((526, 727))
INFO   paperwork.frontend.mainwindow.scan Scan done
INFO   paperwork.frontend.mainwindow.scan Will use tool 'Tesseract'
INFO   paperwork.frontend.mainwindow.scan Animator: Angle 0: (105, 0) (526, 727) -> (73, 20) (222, 323)
INFO   paperwork.frontend.mainwindow.scan Animator: Angle 90: (105, 0) (526, 727) -> (441, 20) (222, 323)
INFO   paperwork.frontend.mainwindow.scan Animator: Angle 180: (105, 0) (526, 727) -> (73, 384) (222, 323)
INFO   paperwork.frontend.mainwindow.scan Animator: Angle 270: (105, 0) (526, 727) -> (441, 384) (222, 323)
INFO   paperwork.frontend.mainwindow.scan Failed to use OCR tool heuristic for orientation detection: [Errno 32] Relais bris? (pipe)
INFO   paperwork.frontend.mainwindow.scan Falling back on Paperwork's heuristic

The tesseract packages installed on my system are:
tesseract 3.04.00-1
tesseract-data-deu 3.02.02-5
tesseract-data-eng 3.02.02-5
tesseract-data-fra 3.02.02-5
and I use “fra” in Paperwork.

[jflesch]

This is more a Pyocr bug than Paperwork but ok. I'll have a look, probably this week-end.

[jflesch]

Got it: It's a regression in Tesseract 3.04 (not present in 3.03).

Since Tesseract 3.x, it's possible to specify "stdin" as input for Tesseract. It's handy since it avoids using temporary files in PyOCR/Paperwork.

Example:

$ tesseract stdin out_file < test.png
Tesseract Open Source OCR Engine v3.04.00 with Leptonica
$ echo $?
0

Unfortunately, with Tesseract 3.04, when trying to detect the orientation of the image, it tries to open the file 'stdin' ... :

$ tesseract -psm 0 stdin out_file < test.png 
Tesseract Open Source OCR Engine v3.04.00 with Leptonica
Error in fopenReadStream: file not found
Error in pixRead: image file not found: stdin
Cannot open input file: stdin
$ echo $?
2

[jflesch]

I'll open a ticket on Tesseract side later. In the meantime, I guess since the argument 'stdin' is not reliable, I'll change PyOCR so it uses a temporary file ... :/

[jflesch]

Confirmed: tesseract-ocr/tesseract#85

[jflesch]

Fixed in PyOCR.

You can run sudo pip install --upgrade pyocr, it should solve this problem.

Please reopen this ticket if you still have a problem.

[tYYGH]

Please forgive the newbie question… What is the proper way to upgrade when the install was done following the instructions there:
https://github.com/jflesch/paperwork/blob/stable/doc/install.devel.markdown#paperwork-in-a-python-virtualenv
Should I run the above command after having run source bin/activate, or not?

[jflesch]

I actually don't work with virtualenv usually, but I think so, yes.

[tYYGH]

I tested. The issue is solved. Thank you.