I'm building a mobile app for photographing documents, what OCR works best on bad images?

For bad document photos on mobile, the “best” OCR usually depends on whether you want:

• highest raw accuracy
• on-device processing
• offline support
• easy mobile SDK integration
• handwriting / form / table support

Best overall for messy document photos

Google ML Kit Text Recognition

• Very good on mobile-captured images
• Works on-device
• Easy to integrate on Android/iOS
• Strong for printed text in imperfect lighting, skew, blur, and perspective issues
• Good choice if you want a practical, fast mobile solution

Best if you want strongest cloud OCR

Google Cloud Vision OCR

• Often better than on-device OCR on very poor images
• Handles rotation, perspective, shadows, and low-quality photos well
• Best when you can upload images to the cloud
• Tradeoff: latency, cost, and privacy concerns

Best enterprise/document OCR

ABBYY FineReader / ABBYY Mobile OCR SDK

• Excellent accuracy for documents
• Strong layout analysis and text extraction
• Good for invoices, forms, IDs, and structured docs
• Usually more expensive / enterprise-oriented

Also worth considering

Microsoft Azure OCR / Document Intelligence

• Very strong for documents and forms
• Good for structured extraction, not just plain text
• Solid cloud option

Tesseract

• Open-source and offline
• Fine for clean images, but usually not the best on bad mobile photos unless you do heavy preprocessing
• Better for research/prototyping than best-in-class mobile OCR

---

What I’d recommend

If you’re building a mobile document-scanning app, a strong practical stack is:

1. Image preprocessing

   • crop document
   • deskew
   • perspective correction
   • denoise
   • increase contrast
   • sharpen lightly

2. OCR engine

   • ML Kit if you want on-device/offline
   • Google Cloud Vision or Azure Document Intelligence if you want better accuracy on tough images

3. Fallback strategy

   • try on-device first
   • if confidence is low, send to cloud OCR

---

Important point

For bad images, OCR quality is often limited more by image capture quality than the OCR engine itself. A good document capture flow helps a lot:

• edge detection
• auto-crop
• live blur detection
• brightness guidance
• auto-capture only when stable

---

Short answer

• Best mobile on-device: Google ML Kit
• Best for worst-quality images: Google Cloud Vision or ABBYY
• Best open-source: Tesseract, but usually not best on bad photos

If you want, I can also give you a side-by-side comparison table or recommend the best OCR for Android vs iPhone vs cross-platform.