Shapecatcher: Draw the Unicode character you want

[dwynings]

http://shapecatcher.com/

[dang]

Some past threads:

Shape Catcher: Find Unicode characters by drawing them - https://news.ycombinator.com/item?id=14284750 - May 2017 (28 comments)

Draw a shape to get a matching unicode-symbol - https://news.ycombinator.com/item?id=11829191 - June 2016 (3 comments)

Shapecatcher: Draw the Unicode character you want - https://news.ycombinator.com/item?id=8403103 - Oct 2014 (38 comments)

Shapecatcher: Draw the Unicode character you want - https://news.ycombinator.com/item?id=5150107 - Feb 2013 (107 comments)

Shapecatcher - draw the unicode character you want - https://news.ycombinator.com/item?id=3222964 - Nov 2011 (63 comments)

Unicode Character Recognition - https://news.ycombinator.com/item?id=3208967 - Nov 2011 (11 comments)

[bobsmooth → dang]

No updates on the twitter account since 2013 unfortunately.

[tbabej]

This reminds me of the now legendary Detexify tool, which performs handwritten symbol recognition for LaTeX [1].

[1] https://detexify.kirelabs.org/classify.html

[sidpatil → tbabej]

Shapecatcher's front page mentions Detexify as well.

[samatman]

Because I have the emotional constitution of a ten-year-old boy, I immediately tried to draw 𓂺 in it, and it didn't find it even after I looked it up and tried to reproduce it as exactly as possible.

Very cool idea anyway, and would probably work for less juvenile use cases.

[Diti → samatman]

I did exactly the same. And I had been expecting the website’s author to have tested it!

[randomstring]

Of course I had to test this by drawing an Interrobang: ‽

It got it right. My first drawing didn't have interrobang at the top of the suggestion list, because I didn't have the top closed. So there is some bias from the font it was trained on.

[tpaschalis → randomstring]

+1. I couldn't get it to understand the partial derivative symbol ∂, unless it was drawn in a very specific way.

But really cool idea, name and execution!

[yreg → tpaschalis]

I just tried to quickly sketch something and got this: Ꭿ

[arc-in-space]

Yeah, not bad https://i.imgur.com/hAi7ren.png

I'm surprised by all the search failures people reported in past threads, just about everything I tried worked decently well

[junon → arc-in-space]

That's in my top five favorite glyphs. It's also called the "seraphim" glyph.

[Causality1]

It would be nice if the suggestions included the keyboard shortcut for inserting that character.

[exporectomy → Causality1]

Depends on the application, doesn't it? It gives the codepoint so you can enter it with, say alt+X in Word or whatever.

[anordal]

Of all the codepoints below 0x100, why is it so bad at Å?

https://en.wikipedia.org/wiki/Å

[exporectomy → anordal]

I found that it won't recognize it with a gap between the A and the circle. Join them together and it gets it as the first choice. Maybe the font(s) it used had them connected?

[sysadm1n]

Funny when you draw a circle and realize how many different variants of the same shape are out there. There's a lot of redundancy in Unicode/UTF8

[not2b]

Cool, but looks like they are excluding Chinese characters.

[joshribakoff]

After a dozen tries it still won’t recognize pi :(

[rcthompson → joshribakoff]

Hmm, being able to narrow down the scope would be useful, e.g. have a series of checkboxes for Latin, Greek, math symbols, etc. (presumably pi would be eligible if you check either Greek or math)

[gizmo686 → joshribakoff]

I can get it to recognize pi fairly consistently if I include a serif on the right leg; although I still haven't been able to get it to rank pi as the first choice.

[TheRealPomax → joshribakoff]

remember that's it's matching against what is basically an SVG image of a specific font's version of pi, not "how people write pi": draw pi, instead of writing pi, and it'll pick it up pretty high up in the list of matches. https://i.imgur.com/k4AnUme.png

[joshribakoff → TheRealPomax]

I drew it curved like this https://upload.wikimedia.org/wikipedia/commons/thumb/0/09/Pi... and it was suggesting Unicode characters comprised entirely of right angles (the suggestions themselves changed each time)

[not2b → joshribakoff]

I got pi as about the seventh choice.

[tasogare]

Doesn't work very well if at all. I tried two times a very simple Chinese character (木) and it was never in the result list. Edit: didn't work with a very simple cuneiform character either.

[mjlee → tasogare]

On the page right now:

> Currently, there are 11817 unicode character glyphs in the database. Japanese, Korean and Chinese characters are currently not supported.

[not2b → mjlee]

The main reason something like this might be useful (since I live near Chinese neighborhoods in the SF Bay Area) would be to recognize Chinese characters, at least those that aren't too difficult to draw. So I hope they can be added.

[TazeTSchnitzel → not2b]

There's many good existing pieces of software specialised in recognising handwritten CJK characters, which understand stroke order and so on. I think the value of this piece of software is that it's for everything else.

[TheRealPomax → not2b]

Why? All OS have IMEs built in, all of which are vastly better at recognizing which character you drew than this can ever be.

[fao_ → not2b]

Google Translate has a handwriting mode, and an image recognition mode.

[tinus_hn → mjlee]

It does recognize Hiragana

[slmjkdbtl → tasogare]

It doesn't recognize any Chinese character for me, maybe not trained with a font containing those characters?