Project 1: Smart Document Scanner with OCR

A mobile-scanner-like app that digitizes documents with automatic edge detection and text extraction.

What You’ll Learn

  1. Edge Detection - Using Canny algorithm to find document edges
  2. Contour Detection - Finding and filtering contours to locate documents
  3. Perspective Transformation - Converting tilted documents to flat, top-down view
  4. Image Enhancement - Adaptive thresholding for clean output
  5. OCR Integration - Extracting text from scanned documents

Key OpenCV Functions

Function Purpose
cv2.Canny() Detect edges in image
cv2.findContours() Find contours from edges
cv2.approxPolyDP() Approximate contour to polygon
cv2.getPerspectiveTransform() Compute transformation matrix
cv2.warpPerspective() Apply perspective transform
cv2.adaptiveThreshold() Enhance document for reading

Usage 

# Run demo with generated sample image
python main.py --demo

# Scan a specific image
python main.py --image /path/to/document.jpg

# Use webcam for real-time scanning
python main.py --camera

# Show debug visualization
python main.py --demo --debug

Algorithm Steps 

1. Load Image
     |
2. Convert to Grayscale
     |
3. Apply Gaussian Blur
     |
4. Canny Edge Detection
     |
5. Find Contours
     |
6. Filter for 4-point contours
     |
7. Order corner points
     |
8. Apply Perspective Transform
     |
9. Enhance with Adaptive Threshold
     |
10. Extract Text (OCR)

Real-World Applications

  • Mobile scanning apps (CamScanner, Adobe Scan)
  • Document digitization systems
  • Receipt scanning for expense tracking
  • ID/passport scanning
  • Whiteboard capture

Code Highlights

Finding Document Corners 

# Approximate contour to polygon
peri = cv2.arcLength(contour, True)
approx = cv2.approxPolyDP(contour, 0.02 * peri, True)

# If 4 points, it's likely a document
if len(approx) == 4:
    doc_contour = approx

Perspective Transform 

# Define destination points (top-down view)
dst = np.array([[0, 0], [width, 0], [width, height], [0, height]])

# Get transformation matrix and apply
M = cv2.getPerspectiveTransform(src_pts, dst)
warped = cv2.warpPerspective(image, M, (width, height))

References