My school graduation project I decided to finish and publish the code. Originally inspired by Windows IME Pad, the idea of the extension is to self-train the net with hand-drawn samples and then use it for quick access to added symbols or phrases.
This was rather an experiment, as I was working with TFJS for the first time. There were also a few previous attempts based on brain.js and HanziLookupJS, but Tensorflow.JS+MN+KNN gives the best recognition quality.
- https://github.com/flowforfrank/classifier (classifier)
- http://jsfiddle.net/z372rv90/ (canvas drawingpad)
- https://www.webtips.dev/writing-your-very-first-neural-network-in-javascript (original project's tutorial)
- https://medium.com/maria-machine/charlie-gerard-machine-learning-for-front-end-developers-with-tensorflowjs-9c908fbfa8ec (TensorFlow.JS + KNN machine learning tutorial)
- https://npm.io/package/@tensorflow-models/knn-classifier (KNN Documentation)
...+ hell ton of DuckDuckGo-ing and Stackoverflow-ing all kinds of stuff :p