I don't think this nifty new project has been mentioned on the list
yet so I wanted to let people know about its existence. This project
basically resuscitates an old project called "Aアあ" which purpose was
to define kanji in SVG format.
I think that the data provided by that project can be useful for our
subject of interest (handwriting recognition ;-) !) too. Here are some
key differences I noticed with the data in Tomoe:
- The strokes are defined in terms of lines and bézier curves. It
should be possible to use these data for training / testing by
- The kanji look more natural. This suggests that they are good
candidates as test data. Recognizers like zinnia can be tuned /
improved so as to maximize the accuracy against this set.
- Information about radicals and stroke components is provided meaning
that the structure information can potentially be used to improve
accuracy or speed up recognition.