Number Plate Recognition System using TrOCR

Built a complete number plate recognition system using Hugging Face's TrOCR — a transformer-based OCR model. The pipeline takes vehicle images, preprocesses them, extracts the number plate region, and runs OCR to read the text accurately.

Evaluation uses CER (Character Error Rate), WER (Word Error Rate), and Levenshtein Distance to measure how accurately the model reads plates compared to ground truth.

This project covers image preprocessing with OpenCV, transformer model inference with PyTorch and Hugging Face Transformers, and result visualization with Matplotlib.

Tech Stack

Python · PyTorch · Hugging Face Transformers · TrOCR · OpenCV · Pillow · Jupyter Notebook

GitHub

https://github.com/devxyasir/Licence-Plate-Detection-Recognition-with-TrOCR