improving OCR to mathematical fonts

[GoogleCodeExporter]

Hi,

what about plugging the detexify feature (see the mentionned website below)
to tesseract in order to enlarge the number of recognized signs and to
offer the capabilities of creating LateX-based source documents : 

http://detexify.kirelabs.org/classify.html

Thanks you

Original issue reported on code.google.com by plutones...@gmail.com on 22 Dec 2009 at 6:33

[GoogleCodeExporter]

or this application maybe : 
http://www.inftyproject.org/en/software.html#InftyReader

Original comment by plutones...@gmail.com on 4 Feb 2010 at 2:35

[GoogleCodeExporter]

Detexify looks like it needs online (drawing) input and therefore won't work 
with images.
Infty doesn't seem to be open source, so that won't be of use.
I am open to hosting an intern to work on this topic.

Original comment by theraysm...@gmail.com on 20 May 2010 at 4:32

• Changed state: Accepted
• Added labels: Priority-Low, Type-Enhancement

[GoogleCodeExporter]

yes, you are correct, Detexify will work on splines, from the drawing. Anyway, a
first step would be to approximate images with splines and subsequently use the
detexify engine.

Concerning the current Google strategy to scan documents and put them on-line, 
it
could be much more efficient to have a real light electronic version (with 
vector
format fonts) instead of an heavy poorly scanned document (with raster 
objects). In
this regard, developing a strategy capable of reconstructing Latex Sources from 
a
scanned scientific document could be very powerful.

Original comment by plutones...@gmail.com on 27 May 2010 at 2:13

[GoogleCodeExporter]

I'm try to train Tesseract for recognizing PDF-images as Latex-Code.
I think the the line-interpretation will make it a bit complicated with 
formulas that are not only one line. So as example a \frac{a}{b} could also be 
an underlined text.
I hope I can find some pattern in the recognized text documents, so that can 
post process them somehow.
It shouldn't be so difficult, because i work with PDFs that are latex-generated.

If you have a suggestion to my plans, please share, thanks.

Original comment by jammi.e...@gmail.com on 20 Apr 2012 at 1:40

[GoogleCodeExporter]

Hi,

My Goal: recognizing PDF-images as Latex-Code
So my input is clean and not rotated. So it seems to be an easy task. Following 
that, I want to tell Tesseract that every black dot is a Symbol/Letter - there 
is no noise.
Is there a easy way to do that? Or do I have to dig in the code?

Also it would be interesting, whether Tesseract recognises overlapping boxes 
(in the box-file) so that a mathematical root would be recognised, but the 
stuff under the root line will be recognised independently.

I will send my progress. Hope I'm right to post here.
Thanks.

Original comment by jammi.e...@gmail.com on 9 May 2012 at 3:24

[GoogleCodeExporter]

[deleted comment]

[GoogleCodeExporter]

What's the strategy then? Report a bug or request a new feature?

Original comment by plutones...@gmail.com on 10 May 2012 at 7:04

[GoogleCodeExporter]

Actually I don't want to report a bug.
Maybe an issue about recognising subscript and superscript?
I want to recognise formulas and symbols that are not in UTF-8, but in Latex.
Tesseract is not build for that, but I want to improve it a bit in that 
direction - by training.

To follow my last comment, I just don't know enough about the training process, 
to use it wisely.

Greetings

Original comment by jammi.e...@gmail.com on 14 May 2012 at 3:21

Attachments:

• picturetoLaTeX.png

[GoogleCodeExporter]

Is there any new advance with this issue? I am interested on scanning 
handwriting notes with math equations and transform them to a LaTeX file. 

Original comment by maikol.s...@gmail.com on 12 Jan 2015 at 4:10

[GoogleCodeExporter]

Issue 1372 has been merged into this issue.

Original comment by zde...@gmail.com on 12 Apr 2015 at 3:06