Need OCR that handles skewed scans and mobile photos

For skewed scans and mobile photos, you want an OCR pipeline that includes image preprocessing before recognition.

Best approach

Use OCR with these steps:

1. Detect orientation / rotation

   • Auto-rotate 90/180/270°
   • Deskew small angles

2. Perspective correction

   • Important for mobile photos taken at an angle
   • Rectifies trapezoid distortion

3. Image enhancement

   • Denoise
   • Increase contrast
   • Adaptive thresholding / binarization
   • Sharpen text regions

4. Text detection + OCR

   • Detect lines/words/blocks first, then run OCR
   • Better than raw OCR on the full image

Good OCR options

Open-source

• PaddleOCR
  • Very strong on real-world images
  • Handles rotated, skewed, and phone photos well
• Tesseract
  • Good for clean scans
  • Needs preprocessing for best results on skewed images
• EasyOCR
  • Simple to use
  • Decent on photos, less robust than PaddleOCR in many cases

Commercial / cloud

• Google Cloud Vision OCR
• Azure AI Vision OCR
• AWS Textract These usually perform well on mobile photos and skewed documents with less tuning.

Recommended choice

If you want open-source and robust: PaddleOCR + preprocessing
If you want best out-of-the-box quality: Google Vision or Azure OCR

If you’re building it yourself

A strong pipeline is:

• Image input
• Orientation detection
• Perspective correction
• Deskew
• Contrast enhancement
• OCR

If you want, I can give you:

1. a Python example for skew correction + OCR, or
2. a recommended library comparison for your exact use case.