OCR Medical Model

Overview

This module implements a Custom Convolutional Neural Network (CNN) for recognizing handwritten medicine names from doctor prescriptions. Unlike generic OCR tools, this model is specifically trained on the Doctors Handwritten Prescription BD Dataset to classify prescription text into 78 distinct medicine classes.

---

Features

• Custom CNN Architecture: A deep learning model built with TensorFlow/Keras for image classification.
• Specific Domain Training: Trained on real-world handwritten medical prescriptions.
• Preprocessing Pipeline: Automated image resizing (64x64), normalization, and label encoding.
• Comprehensive Evaluation: Includes accuracy metrics, confusion matrices, and classification reports.
• EasyOCR Integration: Includes easyocr for potential auxiliary text extraction tasks.

---

Folder Structure

OCR_Medical_Model/
│── data/                      # Dataset directory (Training/Testing images & labels)
│── ocr/
│   │── ocr_script.ipynb       # Main Jupyter Notebook for training, testing & evaluation
│   │── class_labels.json      # JSON file containing the 78 medicine class labels
│   │── sample_images/         # Example prescription images
│   └── models/                # Directory where trained models (.h5, .keras) are saved
│── requirements.txt           # Python dependencies
└── README.md                  # Project documentation

---

Tech Stack

• Deep Learning: TensorFlow, Keras
• Computer Vision: OpenCV, EasyOCR
• Data Handling: Pandas, NumPy
• Evaluation: Scikit-learn (Confusion Matrix, Classification Report)
• Visualization: Matplotlib

---

How It Works

1. Data Loading: Reads images and their corresponding labels from the dataset CSVs.
2. Preprocessing:
   • Images are resized to 64x64 pixels.
   • Pixel values are normalized to the [0, 1] range.
   • Labels are encoded using LabelEncoder and converted to categorical vectors.
3. Model Architecture:
   • Conv2D Layers: Extract features from images.
   • MaxPooling2D: Reduces spatial dimensions.
   • Dense Layers: Fully connected layers for classification.
   • Dropout: Prevents overfitting.
   • Output Layer: Softmax activation with 78 units (one for each medicine class).
4. Training: The model is trained for 20 epochs using the Adam optimizer.
5. Evaluation: The model is tested on a separate test set to ensure generalization.

---

Installation

1. Clone the repository:

   git clone https://github.com/SanjayKumar3110/OCR_Medical_Model.git

2. Install dependencies:

   pip install -r requirements.txt

   Note: Ensure you have a Python environment (recommended 3.10+) set up.

---

Usage

Training and Testing the Model

1. Navigate to the ocr directory:

   cd ocr

2. Open the Jupyter Notebook:

   jupyter notebook ocr_script.ipynb

3. Run the cells sequentially to:
   • Load and preprocess the data.
   • Train the CNN model.
   • Evaluate performance on the test set.
   • Save the trained model to the models/ directory.

---

Dataset

The dataset used for training is publicly available on Kaggle:

The project uses the Doctors Handwritten Prescription BD Dataset from Kaggle, which contains:

• Training Data: Images of handwritten medical words.
• Labels: CSV files mapping images to medicine names.

---

Future Work

• Data Augmentation: To improve model robustness against variations in handwriting.
• Hyperparameter Tuning: Optimizing learning rates and batch sizes for better accuracy.
• Integration: Connecting this module with the main MediScan backend for real-time predictions.
• Transformer Models: Experimenting with Vision Transformers (ViT) for potentially higher accuracy.