Hacker News Reader

Comment by gamegoblin | Hacker News Reader

Shazam for fonts. Point your phone camera at some text and it tells you the top N fonts it thinks it is with C% confidence.

Niche market for graphic designers.

There's the website What The Font [1], but it's not a super user friendly app like Shazam.

I think the hardest part of this would be the huge amount of time it'd take to get a good dataset and train whatever machine learning model you'd probably wind up using.

[1] https://www.myfonts.com/WhatTheFont/

View on HN

Nice idea. I think you will find the hardest thing will be that so many typefaces are near clones of each other. It might be easier to just match to the closest open source typeface.

Definitely true. That's why you'd have to print out the top N fonts it thinks it's similar to.

Something like

    1. Foo Bar Lite Italic - 99.1%
    2. Foo Bar Italic - 97.2%
    3. Baz Bar Lite Italic - 97.0%
    4. Foo Baz Italic - 95.4%
    ... etc maybe give the top 10 or so ...

Another problem is many logos and whatnot have been hand-tweaked from their original font.

Yes it is a hard problem. It might be worth trying on a small scale with a few close typefaces to see how hard it is to separate them and what sort of accuracy could be achieved.

Definately not a hard problem. For logos, impossible because of customisation, but for regular text, it's simple. Capture image, make a box around text, warp to flatten, highten contrast, split characters, process characters for features, check against database of known features of fonts. It's similar to a facial recognition algorithm.

Problem is no monetization potential, so it's hobby project level.

I think you are under estimating how similar many typefaces are to each other. When even human experts have a hard time determining which typeface is which you know it is a hard problem.

I do agree 100% with you that any such product would be impossible to monetarise. I would certainly like to use it so I hope someone makes it.

> When even human experts have a hard time determining which typeface is which you know it is a hard problem

A font is just a series of images with standardised traits. Computers are great at recognising these traits, and perfect for cross comparing them. It's just pixels.

dataset could be provided through google fonts maybe. you'd need to replicate a few iphone camera quality versions for each font though.

any links for machine learning images? I could give this a go.

Unfortunately, google fonts is far too small a font set to be useful for a hardcore graphic designer.

It would be useful for "Here is a free font on google fonts that looks similar to whatever you are looking at".

Reply on news.ycombinator.com