They didn't think to weight the prior probabilities by usage frequency* - drawing a reasonable ? gives me ȓ, ᕉ, ╔, ᣑ, Ѓ, ק, ᒌ, ŕ, ᒤ, ᒦ, ņ, ᒯ, ѓ, and finally ?.
I'm also guessing that they're directly comparing the handwritten character to some version of the unicode character rather than with human attempts to draw the character. Human drawings are often quite different (more slanted, stylised etc.) than typeface characters. This is much more forgiveable though because assembling a good dataset for human drawn characters is hard (especially for any reasonable chunk of the unicode set).
(*this is fairly easy to do: just find some large source of typical unicode, like Wikipedia in all languages, and index them).
Some other useful unicode websites:
http://www.fileformat.info/info/unicode/char/search.htm
http://www.fileformat.info/info/unicode/block/index.htm
(I've added the above as search engines in Chrome with short mnemonics for the keyword.)
There's also:
http://unicode.johnholtripley.co.uk/ -- mobile unicode support tables
http://apps.timwhitlock.info/emoji/tables/unicode -- emoji :D
http://panmental.de/symbols/info.htm
and of course:
http://copypastecharacter.com/
Want more unicode resources? There's a list of other resources here:
http://joewlarson.com/blog/2014/01/01/useful-unicode-resourc...
This is great, it's like http://detexify.kirelabs.org/classify.html but for unicode instead of latex.
An interesting side effect of this is that it shows once again why (naive implementations of) international domain names presented such a large security risk. Just draw an "A" and look at the results...
There is also http://www.nciku.com/ for chinese characters.
It seems (and is logical, I suppose) that you have to match the symbol pretty closely to get what you're looking for. I drew a car twice, the first time getting absolutely nothing relevant, and the second time -- trying to be more precise and using all the space available -- got automobile, taxi, bus, etc, etc.
edit: The primary problem here, I mean, is that if you don't know what the symbol looks like and want to see if it exists, you might not get hits the first time you try to draw, but it might not actually exist anyway.
It doesn't support every unicode character yet, but it's getting there. For example, it recognized the Kannada character ttha (ಠ), but it doesn't know the poop (💩) character.
I've often thought that the best way to get access to the richness of Unicode would be a drawing pad, perhaps as part of the keyboard, or as an on-screen area, for use with a mouse. Character maps just seem clumsy to me.
For fun, I tried Eth ð, Thorn þ, and Hungarian ű, all of which it got, but not as the first choice. It did not find the Ing rune, which looks a bit like a < and > combined.
Oh wow, the drawing mechanism is really satisfying.
Too bad about not supporting æ¼¢å—. The only half-decent IME pad is on Windows. Online ones (kanji.sljfaq.org) and Xorg ones (ibus-mozc) are just horrendously bad at detection. I usually have to resort to multi-radical lookups.
Nice. I wish the GNOME Character Map¹ could do this.
I drew #. Returned a bunch of characters but none of them was regular old 0x0023.
Took me several tries to get # (regular ascii number sign - i.e. shift+3).
This is pretty cool. I expect that even if the code is open source, that the real value is in the dataset used. Does anyone know the licensing information ?
This is one of the first sites I learned about from HN a few years ago. Extremely useful.
I just did an 8 and got nothing close, but wow there are a lot of interesting Unicode characters.
It works. Good work.
i drew a penis and got the "male sign" :-)
What would perhaps be a worthwhile feature is the ability to input already known unicode names into a text box after seeing the results, to feed the database with more useful matches.
For example, tried drawing a somewhat joined 'TM' a couple times but no matches for 'Trademark symbol', however a way to manually input that unicode/name might provide the database with a positive match for the next user trying to find it.