[prev in list] [next in list] [prev in thread] [next in thread] 

List:       kde-devel
Subject:    GSoC idea: improving scanning and OCR in KDE (skanlite/kooka)
From:       José_Manuel_Santamaría_Lema <panfaust () gmail ! com>
Date:       2012-03-06 13:00:57
Message-ID: 201203061401.24847.panfaust () gmail ! com
[Download RAW message or body]

[Attachment #2 (multipart/signed)]


Hi,

I'm considering to apply to GSoC this year, and if I do, I would like to 
improve the status of scanning and optical character recognition in KDE; being 
more specific:


What I want to achieve
--------------------------------

A few years ago I had to study electronics stuff at my university following 
class notes only available in paper. I was annoyed because I couldn't use 
ctrl+F with a paper, so I investigated a bit about OCR stuff and I found the 
open djvu file format[1].

So I tried to produce a djvu document with KDE and my free operating system; 
it was (very difficult|impossible); if recall correctly I did something like 
this: I tried to figure out a workflow scanning a couple of pages as 2 jpeg 
files, then I tried to join them in a djvu multipage document using shell 
commands, and suceeded. However I couldn't find out how to do the OCR part, 
iirc I tried a couple of free ocr programs (I didn't tried ocropus; I don't 
remember if either that program didn't exist at that moment or I just didn't 
know about it) but their output was just the text without the coordinates 
where the texts are located, which would be needed to produce a proper text 
layer in the djvu document.

So I gave up and rebooted on Windows and I used a propietary software to 
produce the document; it worked quite well, I just fed the papers in my 
scanning device and produced a multipage document; when done I just clicked a 
menu item labelled as something "process the document using OCR" and that's 
it. I don't remember very well the name of the software I used, but I'd swear 
it was "Document Express"[2].
The result was excellent, and you can download the produced document here:
http://alioth.debian.org/~santa-guest/gsoc2012/apuntes_te.djvu
As you can see, the size of the document is reasonable (only 2.4M) and you can 
do ctrlf+F "zener" and read stuff about zener diodes.

So... to sum up: it was/is easier to produce good djvu documents with 
propietary software. I want a KDE'ish program to replace the expensive 
"Document Express".


Some technical details
--------------------------------

Currently we have a couple of KDE programs to scan documents: skanlite and 
kooka. skanlite is quite simple (doesn't do OCR stuff), uses the modern liksane 
library, it's in extragear and works fine. kooka provided more functionality in 
the KDE 3 old days than skanlite today (seems it was able to do some basic OCR 
stuff), uses its obsolete libkscan library, it's in playground and I don't know 
if it works or not because I don't have an scanning device right now, but at 
least it builds properly.

So... looks like the tasks to do to achive my goal would be:
1. If needed, extend libksane functionality in order to make it a good 
replacement for the old libkscan.
2. Port kooka to the modern libksane.
3. Add ocropus support to kooka (I heard with ocropus you can get the 
coordinates of the texts, but I don't know for sure yet)
4. Code something in kooka to produce djvu documents.


[1]http://en.wikipedia.org/wiki/DjVu
[2]https://www.caminova.net/en/shop/item.aspx?itemid=3

["signature.asc" (application/pgp-signature)]

>> Visit http://mail.kde.org/mailman/listinfo/kde-devel#unsub to unsubscribe <<


[prev in list] [next in list] [prev in thread] [next in thread]