On Block Recognition
July 19, 2024 | backboth Palm OS and Windows Mobile have support for Graffiti characters through something WinMobile calls the "block recognizer" - i was wondering if that would be easy to implement nowadays?
the first thing most people do when learning to do ML is make an MNIST classifier - what if we just expand that to allow typing every character in a given set? just train it on a few thousand instances of special characters, letters, numbers, etc. in every language you want. need japanese input? draw up examples of hiragana/katakana (kanji would be a stretch just because of the sheer amount of data needed). want accents? draw up (character list) with every single one you need. want a dedicated amongus key? spit out unicode u+d9e. things like that
i think about this because of an upcoming handheld computer project i'd like to work on. i'm of the belief that any computing enthusiast should be making some sort of custom device at some point, like a cyberdeck, and i personally would love a little handheld Linux terminal with block recognition.