🧠 OCR Evaluation Tool

This project is a comprehensive OCR (Optical Character Recognition) Evaluation Framework designed to rigorously test and analyze the performance of custom OCR pipelines. It supports printed and handwritten text recognition, and delivers detailed performance insights including character-level metrics, word accuracy, and confusion matrices.

---

📁 Project Structure

• dataset/img/ – Folder containing input images for OCR evaluation
• dataset/text/ – Corresponding ground truth .txt files
• backend.py – Includes extract_text() for OCR inference and preprocess_image() for image preparation
• evaluate.py – Main evaluation script computing metrics and visualizations

---

📊 Metrics Computed

• ✅ Word-Level Accuracy
• 🔠 Character-Level Accuracy
• ✏️ Average Levenshtein Distance
• 🎯 Precision, Recall, and F1 Score (char-wise)
• 🔍 Character Confusion Matrix via Seaborn heatmap

---

🛠️ How It Works

For each image in dataset/img/, the following is performed:

1. Ground truth is loaded from the corresponding .txt file.
2. The image is preprocessed using preprocess_image.
3. Text is extracted using a custom OCR engine via extract_text.
4. Evaluation metrics are computed:
   • Word-wise comparison
   • Character-wise match count
   • Levenshtein distance
   • Confusion matrix data

Results are accumulated and visualized in real-time.

---

🖥️ Sample Output (Console)