OCR using cuneiform does not work

[FriedChicken]

Binary package hint: gscan2pdf

The cuneiform version in ubuntu has no libmagick++-support (Bug #654767). Therefore it can only process uncompressed BMP v3 images.

Trying anything else leads to OCR being cancelled. On the console this is printed:
> /tmp/lD1hIbHasU/ThgzjU3Pqw.pnm is not a BMP file.
> *** unhandled exception in callback:
> *** Error: cannot open /tmp/lD1hIbHasU/DWr2n3tweG.txt
> *** ignoring at /usr/bin/gscan2pdf line 12513.

As a workaround gscan2pdf should convert the images before passing them over to cuneiform.

ProblemType: Bug
DistroRelease: Ubuntu 10.10
Package: gscan2pdf 0.9.31-2
ProcVersionSignature: Ubuntu 2.6.35-22.33-generic 2.6.35.4
Uname: Linux 2.6.35-22-generic x86_64
NonfreeKernelModules: fglrx
Architecture: amd64
Date: Mon Oct 4 20:51:24 2010
InstallationMedia: Kubuntu 10.04 LTS "Lucid Lynx" - Release amd64 (20100427)
PackageArchitecture: all
ProcEnviron:
 LANGUAGE=
 LANG=de_DE.utf8
 SHELL=/bin/bash
SourcePackage: gscan2pdf

[Jeffrey Ratcliffe (jeffreyratcliffe)]

This patch fixes things for me

[FriedChicken (domlyons)]

Thank you! Yes, this should work.

[FriedChicken (domlyons)]

I'm not sure if I should file a new bug or append it to this one...

Now cuneiform is built with libmagick++-support (Bug #654767 fixed for maverick, fix for Lucid is in proposed repository). So cuneiform can perform nearly any image format. But OCR with cuneiform still doesn't work: The OCR tab simply stays clear.

Manually starting cuneiform an a scanned image in /tmp/CrazyFolderName/RandomImageName.pnm works and shows a pretty exact result. So it's not a problem of cuneiform or a unusable scanned document.

[FriedChicken (domlyons)]

Replacing the "-f hocr" option for cuneiform by "-f smarttext" solved this. (Instead of "smarttext" "text" should work, too. But smarttext is probably better in most cases.)

Did "-f hocr" work for anyone at all?

[Jeffrey Ratcliffe (jeffreyratcliffe)]

I need to do some more work on that patch.

It seems that the hocr output from this version of cuneiform is a box
per letter, which gets the letters in the right place, but is useless
for searching.

Can anyone check cuneiform 1.0.0 to see if it the same there?
Otherwise, I'll probably switch to plain text.

[Jeffrey Ratcliffe (jeffreyratcliffe)]

It works fine with cuneiform 1.0.0

[Launchpad Janitor (janitor)]

[Expired for gscan2pdf (Ubuntu) because there has been no activity for 60 days.]

[Marja Erwin (marja-e)]

I am running into the same bug.

[Marja Erwin (marja-e)]

I marked this as new, because it wasn't showing up when I searched for gscan2pdf.

[Jeffrey Ratcliffe (jeffreyratcliffe)]

This is fixed in gscan2pdf 1.0.0