LiveScribe is an experimental computer vision project for recognizing handwritten math expressions from a drawing canvas and evaluating them.
I built this to explore how far I could get with symbol recognition, preprocessing, and expression parsing using deep learning, while keeping everything interactive and visual.
The project uses a pretrained transformer-based handwriting model for character-level recognition, combined with custom OpenCV preprocessing to clean and normalize user drawings before inference.
- Drawing-based handwritten input
- Image preprocessing (thresholding, resizing, normalization)
- GPU-accelerated inference with PyTorch
- Model configs and tokenizer included (weights excluded)
Python · PyTorch · OpenCV · NumPy · Gradio
pip install -r requirements.txt