Artificial-Intelligence-Final-Project-Handwriting-Recognition

This repository now includes an end-to-end handwriting processing pipeline that:

1. Takes an input image of handwritten text.
2. Runs OCR to extract letter/word sequences.
3. Applies word prediction/correction for noisy OCR output.
4. Exports the predicted text to both .txt and .docx.

Quick start

Create and activate/install with your virtualenv (or use .venv/bin/... commands directly):

sudo apt-get update
sudo apt-get install -y python3.12-venv tesseract-ocr
python3 -m venv .venv
.venv/bin/python -m pip install --upgrade pip
.venv/bin/pip install -r requirements.txt

Run the pipeline:

.venv/bin/python handwriting_pipeline.py \
  --image path/to/handwritten_image.png \
  --output-stem outputs/predicted_words

Outputs:

• outputs/predicted_words.txt
• outputs/predicted_words.docx

Train and use the CharacterCNN model

Train the CNN and save a checkpoint:

EPOCHS=1 MAX_TRAIN_SAMPLES=512 MAX_VAL_SAMPLES=256 \
  .venv/bin/python Letter_Detection.py

Default checkpoint path:

• models/character_cnn.pt

Run pipeline with CNN-assisted refinement:

.venv/bin/python handwriting_pipeline.py \
  --image path/to/handwritten_image.png \
  --output-stem outputs/predicted_words \
  --cnn-checkpoint models/character_cnn.pt

Disable CNN and use OCR-only mode:

.venv/bin/python handwriting_pipeline.py \
  --image path/to/handwritten_image.png \
  --output-stem outputs/predicted_words \
  --disable-cnn

Demo

You can generate a sample handwritten-style image and run the full pipeline:

.venv/bin/python demo_pipeline.py
.venv/bin/python handwriting_pipeline.py \
  --image sample_handwriting.png \
  --output-stem demo_output/predicted_words

This creates:

• sample_handwriting.png
• demo_output/predicted_words.txt
• demo_output/predicted_words.docx

Accuracy check

Run a quick evaluation against the built-in demo reference text:

.venv/bin/python evaluate_pipeline.py

This prints the expected text, predicted text, and word-level accuracy.