invoice OCR bad scan errors

Invoice OCR can fail in predictable ways, especially on bad scans. Common error types include:

• Blurry text / motion blur: characters merge or disappear
• Low contrast: faint print, faded ink, shadows
• Skewed or rotated pages: lines and tables no longer align
• Cut-off edges: missing totals, invoice numbers, tax fields
• Cramped layouts / tables: columns get mixed up
• Noise and artifacts: speckles, streaks, scanner lines
• Poor image resolution: small fonts become unreadable
• Handwriting or stamps: often misread as printed text
• Mixed fonts / logos / watermarks: confuse extraction
• Compression artifacts: from JPG/PDF exports

Typical OCR mistakes on invoices:

• 0 vs O, 1 vs I vs l
• 5 vs S, 8 vs B
• Misreading currency symbols or decimal points
• Wrong dates, invoice numbers, PO numbers
• Table row shifts causing line-item quantities/prices to mismatch
• Dropped totals, tax, or vendor address

Ways to reduce errors:

• Scan at 300 DPI or higher
• Use grayscale or black-and-white with good contrast
• Deskew and crop the page
• Remove noise and sharpen lightly
• Prefer PDF text extraction when available instead of OCR
• Use field validation:
  • invoice total = sum of line items + tax
  • date format checks
  • vendor name match
• Use template-based extraction for recurring vendors
• Keep a human review step for low-confidence fields

If you want, I can also help with:

1. a list of OCR error categories for invoices,
2. preprocessing steps to improve scans, or
3. validation rules to catch bad OCR output automatically.