Projects / Computer Science / Computer Science project 8

Handwritten Hindi Character Recognition

Approx. component budget: Rs. 0 - 1,000

Objective

Train a CNN on a Devanagari handwriting dataset and make it work on characters drawn by different people.

What you will build

A web page with a drawing box where the user writes a Devanagari character and the model shows the predicted character with its top 3 guesses.

Results you can expect

• Test accuracy and top-3 accuracy
• Confusion between look-alike characters
• Effect of data augmentation
• And accuracy on 200 samples written by classmates

What you will learn

CNN design, augmentation, confusion analysis, handling of regional scripts, and exporting a model to run in the browser.

Tools and notes

Python, TensorFlow/Keras, Devanagari Handwritten Character Dataset, TensorFlow.js, HTML canvas.

Resume points from this project

• Built and tested Handwritten Hindi Character Recognition; objective: train a CNN on a Devanagari handwriting dataset and make it work on characters drawn by different people.
• Measured and analysed test accuracy and top-3 accuracy and confusion between look-alike characters; wrote the report and presented the results in a viva. Add your own numbers: [x].

Replace [x] with your own measured results. Write only what you really did.

More computer science projects

Back to all projects