Machine learning lets micro decode your handwriting

This rig will take the letters you write on the touchpad using a stylus and turn them into digital characters. The system is very fast and displays near-perfect recognition. This is all thanks to a large data set that was gathered through machine learning.

The ATmega644 that powers the system just doesn’t have the speed and horsepower necessary to reliably recognize handwriting on its own. But provide it with a dataset to compare against and you’re in business. [Justin] and [Stephen] designed a neural network algorithm that took a large volume of character handwriting samples, and boiled them down into a set of correlations that can be referenced when encountering a new entry. This set is about 88 kilobytes, too much to store in the microprocessor, but easy to reference from an external flash memory device.

There’s plenty of gritty details in the write up linked above, but you may want to start with the video overview found after the break.

http://www.youtube.com/watch?v=z7HwuC4YcKA

Comments

  1. Willrandship says:

    So, homebrew scribblenauts? Awesome.

  2. Victor Aprea says:

    W00t! ECE476 / Cornell ECE represent! Really cool, way to go guys!

  3. plfx says:

    “stop talking please”
    harsh

  4. Robot says:

    I love hard ass professors.

  5. Zee says:

    Harsh? Hardass? That’s not being a hardass? What kind of schools did you people go to?

    • Robot says:

      Fair enough; it’s just that the professor’s voice and demeanor conjured up an EE professor I had who was soft spoken and unfailing polite but could cut you down with a word. Lecture started 5 min. early and went to the bell; he would write on the board with one hand while erasing with the other. What made him a hard ass was that he expected the students to be serious, pushed us to the limits of our intelligence and cut us no slack. It was exactly what I didn’t know that I needed at the time.

      – Robot

  6. rue_mohr says:

    ah just what I was looking for, now if I can just feed it an array of fourier audio data I can have an avr that can recognize voice commands.

  7. Jorge says:

    I’m spaniard and ECES is phonetically equal to heces, which is feces in english. Cool project, not sure if cool name…

  8. nice! ECE projects are always great.

Speak Your Mind

*

Related Hacks in news

  • Graphic equalizer display flashes LED sign to the beat
  • Hackaday around the web and into the future
  • The making of a Vacuum Tube
  • Amphi-Cycle lets you ride the trails, the waves, and back again
  • Heathkit closes down, again.