Tegaki 0.3 released
Mathieu Blondel
Tegaki is a free and open-source handwriting recognition project for
Chinese and Japanese characters.
Tegaki:
* is free and open-source
* is multi-platform
* focuses on Chinese (simplified and traditional) and Japanese characters
* supports 2 different recognition engines
* aspires to work on both desktop-PCs and mobile devices
*** Highlights for this release
- New recognizer "Wagomu"
- Traditional Chinese handwriting model
- Character rendering tool with support for PNG, SVG, PDF and animated GIF
- Windows and OS X bundles
*** Contributors for this release
- Mathieu Blondel
- Christoph Burgmer
- Ian Jonhson
- Roger Braun
- Conrad Parker
*** Homepage
Cheers,
The Tegaki project contributors
Christoph Burgmer
That's pretty many features for one release.
-Christoph
Ian Johnson
It's great to see more and more activity and contributors too!
Ian
Ian Johnson
Roger Braun
What sources do you use for the models? Tomoe? You can use the model I
made that combine KVG with Tomoe data, as KVG is now under a
CC-Attribution-Share Alike license, so it could even be put into
distributions. My models for zinnia are available from
http://github.com/rogerbraun/KVG-Tools/downloads. They have much
better match1 accuracy.
"Japanese"
match1
Accuracy/Recall: 76.00
Precision: 88.00
F1 score: 81.56
"KVG-Tomoe"
match1
Accuracy/Recall: 91.00
Precision: 96.00
F1 score: 93.43
--
Roger Braun
http://yononaka.de
roger...@student.uni-tuebingen.de
Mathieu Blondel
> What sources do you use for the models? Tomoe?
Yes, I used Tomoe for now. I was waiting for the Python version of
your tool and I'm also waiting for the git repo of KanjiVG. This will
make it possible to include it in the Tegaki git repo as a submodule.
I could have added your Ruby tool and the KanjiVG file directly in the
repo but I didn't want to make the repo any bigger than it is already.
Plus, like Christoph said, there are already many new features, we
need to keep some for 0.4 too :-)
BTW, one can create a shallow clone of repo with the --depth option.
This way all the history isn't downloaded, only the last few
revisions. But a shallow clone can't be cloned or fetched from.
If you need details as to how models are prepared, you can have a look
at tegaki-models/Makefile.
> You can use the model I
> made that combine KVG with Tomoe data, as KVG is now under a
> CC-Attribution-Share Alike license, so it could even be put into
> distributions. My models for zinnia are available from
> http://github.com/rogerbraun/KVG-Tools/downloads.
Thanks, I added a link on tegaki.org
Mathieu