Pen-based Handwritten Character Recognition for Kannada Numbers

This was my final project during my undergrad as an ECE student at Manipal, India, where I tackled the challenge of recognizing handwritten Kannada numbers on smartphones.

The motivation behind this project was that languages like Kannada, a local language from southern India, have a vast number of characters and combinations, making qwerty keyboard input impractical.

For a reliable, real-time character input, online handwriting recognition offers a better solution. To clarify, online handwriting recognition processes the coordinates of character strokes using pattern recognition techniques (rather than relying on image-based recognition). This approach is significantly faster and more efficient for real-time use.

In this project, I developed an Android app to collect handwriting data for Kannada numbers. I implemented preprocessing steps such as smoothing, resampling, and normalization to ensure consistent data quality. Finally, I investigated the effect of feature reduction and data augmentation using autoencoders on the performance of classification algorithms like the SVM.

This work culminated in a short oral paper, which I had the opportunity to present at ICCMEH '23. You can read the short paper here.