Re: Math / equation detection module for Tesseract 3.02

[Igor]

I have tried to OCR this simple image using math module whith this command:

tesseract.exe simple.bmp output -l equ

↕↕⊣⊳↗∫

But as you can see output is garbage.

When I use tesseract with english dictionary I get better result

tesseract.exe simple.bmp output -l eng

11+X

Is it possible to OCR fractions which span over several lines using math module or this math module is math optimiziation for in-line math symbols?

[Matan Safriel]

Same here (on linux in my case).

Version 3.03.

Also tried dropping in the latest online downloadable equations add-on's equ.traineddata, which might not be compatible. With and without it, same results.

Errors during the run:

Bad properties for index 1, char ⊒: 255,0 255,0 0,32767 0,32767 0,32767

Bad properties for index 2, char ⊟: 255,0 255,0 0,32767 0,32767 0,32767

Bad properties for index 3, char ⇓: 255,0 255,0 0,32767 0,32767 0,32767

Bad properties for index 4, char ≆: 255,0 255,0 0,32767 0,32767 0,32767

Bad properties for index 5, char ⊅: 255,0 255,0 0,32767 0,32767 0,32767

Bad properties for index 6, char ↧: 255,0 255,0 0,32767 0,32767 0,32767

Bad properties for index 7, char ⇥: 255,0 255,0 0,32767 0,32767 0,32767

Bad properties for index 8, char ∅: 255,0 255,0 0,32767 0,32767 0,32767

Bad properties for index 9, char ⋕: 255,0 255,0 0,32767 0,32767 0,32767

.

.

.

.

[Tom Morris]

A few notes on equation detection:

- In the 3.02 announcement, the feature is listed as "experimental equation detector" (emphasis added)

- there's no documentation on what's actually in the "equ" trained data file

- the equation detector appears to be turned off by default https://code.google.com/p/tesseract-ocr/source/browse/ccmain/tesseractclass.cpp#486

Note also that the feature is called detection, not recognition, so it's entirely possible it's just something to help out with page segmentation.

Tom

[Tom Morris]

p.s. For anyone interested in the topic, here's a masters thesis on math equation detection and segmentation (and the work was done using Tesseract):

https://vtechworks.lib.vt.edu/handle/10919/46724

https://vtechworks.lib.vt.edu/bitstream/handle/10919/46724/Bruce_JR_T_2014.pdf?sequence=1

[Matan Safriel]

Thanks Tom,

I guess turning the feature on in tesseract might indeed help its performance :-)

As for the thesis, a very thorough, informative, well written and objectively self-evaluated one. From its Table 14, I am not really able to convince myself that plugging its MEDS module into tesseract would perform better than tesseract 3.02 does as is. Table 14 seems to show a trade off between true positives increase and false positive decrease. False discovery rate doubles when using its MEDS module, according to that table. 

Matan

--
You received this message because you are subscribed to a topic in the Google Groups "tesseract-ocr" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/tesseract-ocr/_V7pOll2kPo/unsubscribe.
To unsubscribe from this group and all its topics, send an email to tesseract-oc...@googlegroups.com.
To post to this group, send email to tesser...@googlegroups.com.
Visit this group at http://groups.google.com/group/tesseract-ocr.
To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/0c0aaca9-cd7d-41df-93cc-6b5b6bce559a%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.