To recognize and categorize different mouse gestures using simple logistic regression. The model is trained with very high accuracy (98%). I had some doubt about overfitting too but it works very well in practice.
- U (letter u): U
- Alpha (greek alpha letter): α. Unfortunately, the way I write alpha makes it looks like the Ichthys symbol.
- Hat: ^
- Z (letter z): Z
- The derivative symbol (looks like an intverted 6): ∂
- Triangle (equilateral triangle pointing upward): △
- Greater than: >
This list is definitely extendable simply by adding more training samples and let the model learn them.
To start the demo, start the http server first by running
$python server.py
Then you can run the demo jar
$java -jar Demo.jar
The demo will exit after 20s. To run it for a long while, use the following (Ctrl + C twice to exit)
while true ; do java -jar Demo.jar ; done
Register the drawing by holding down left ctrl, draw a pattern with the mouse, and release left ctrl. The server will try to recognize the pattern and a small popup will appear if a symbol is recognized.
- JNativeHook (for the demo only)
- scikitlearn
- Argo JSON