Document Scanner Project

This repository implements a document scanner in Python as part of the "Artificial Vision" course at URJC. The application takes a photo of a paper document and outputs a rectified (top-down) view of the paper, mimicking the result of a traditional scanner.

Features

• Detects the four corner points of a paper document using Harris keypoints and SIFT descriptors.
• Applies image thresholding and morphological operations to clean up the image.
• Matches keypoints against a reference set to localize corners.
• Performs a perspective transform to generate a scanned version of the document.
• Includes a test mode that batch-processes a folder of images and displays results interactively.

Technologies Used

• Python 3.12
• OpenCV 4.11
• NumPy
• SciPy
• Matplotlib
• Pandas

How to Use

Single Image Mode

python scanner.py path/to/image.jpg

Batch Mode (Test Folder)

python scanner.py --test path/to/test_folder

Reference Data

The system expects a CSV file with manually labeled corner points and a folder with the corresponding reference images:

• learning_corner_points.csv
• muestraTrainTest/learning/

These are used to compute reference descriptors used in corner detection.

Output

• Transformed (rectified) document image.
• In test mode, displays side-by-side visualization of preprocessing and corner detection.

Project Structure

scanner.py                # Main program
learning_corner_points.csv # CSV with labeled corner points
muestraTrainTest/          # Contains training and test images

Requirements

Install the required packages with:

pip install -r requirements.txt

Notes

• Corner detection may fail if document edges are not clearly visible or poorly illuminated.