TABLE EXTRACTION using OCR

This project automates table extraction from scanned images or documents using a hybrid deep learning approach:

1. Table detection with Hugging Face’s Table Transformer.
2. Cropping & padding detected regions.
3. Table structure recognition with PaddleOCR PPStructureV3.
4. Export to Excel for easy downstream use.

---

🚀 Features

• Detects multiple tables in a single page image.
• Automatically crops detected regions with margin padding.
• Recognizes table structure and cell text.
• Exports structured tables into Excel (.xlsx).
• Supports both CPU and GPU acceleration.
• Works with JPEG / PNG / scanned document images.

---

📂 Project Structure

project/
│── main.py              # Main pipeline script
│── requirements.txt     # Python dependencies
│── README.md            # Documentation
│── input/               # Place input images here
│── output/              # Cropped tables + extracted Excel files

---

⚙️ Installation

1. Clone the repository

git clone https://github.com/yourusername/table-extraction.git
cd table-extraction

2. Create a virtual environment (recommended)

python3 -m venv venv
source venv/bin/activate   # Linux / Mac
venv\Scripts\activate      # Windows

3. Install dependencies

pip install -r requirements.txt

---

▶️ Usage

1. Place your input image (e.g., try.jpeg) in the input/ folder.

2. Run the script:

python main.py

3. Outputs:

• Cropped tables → stored as .png in output/.
• Extracted structured tables → saved as Excel .xlsx in output/.

---

🔄 Workflow / Userflow

1. Input Image

   • User provides a scanned image/document (JPEG or PNG).

2. Table Detection

   • Hugging Face Table Transformer locates tables and returns bounding boxes.

3. Cropping & Expansion

   • Each bounding box is expanded by 0.3 cm margin (converted to pixels).
   • The table region is cropped and saved as an image.

4. Table Recognition

   • Cropped tables are passed to PaddleOCR PPStructureV3.

   • Recognizes:

     • Table grid structure (rows/columns).
     • Text inside each cell.

5. Export Results

   • Each detected table is exported as:

     • table_XX.png (cropped table image).
     • table_XX.xlsx (structured Excel file).

6. Logs & Timing

   • Console prints:

     • Number of detected tables.
     • Confidence scores.
     • Output file paths.
     • Total execution time.

---

💻 Example Run

python main.py

Example output:

Table 0 cropped: output/table_00.png (score=0.912)
Table 1 cropped: output/table_01.png (score=0.878)
Parsed results saved in output/
Execution time: 1.42 mins

---

⚡ Tech Stack

• PyTorch → Table Transformer inference.
• Hugging Face Transformers → Pretrained model loading.
• Pillow → Image manipulation.
• PaddleOCR (PPStructureV3) → Table structure + OCR.
• NumPy → Array operations.