High-performance text correction for OCR output using Hyperscan and SymSpell.
Requires Hyperscan system library:
# macOS
brew install vectorscan # ARM
brew install hyperscan # Intel
# Ubuntu/Debian
apt-get install libhyperscan-devThen:
pip install search-and-replacefrom search_and_replace import SpellCorrector, OCRCorrector, PatternCorrector
# Fix common OCR confusions (0→O, 1→l, rn→m)
ocr = OCRCorrector()
ocr.correct("He11o W0rld") # "Hello WOrld"
# Spell correction with bundled dictionary
spell = SpellCorrector()
spell.correct("helo") # "hello"
# Pattern matching with Hyperscan (fast multi-pattern)
patterns = PatternCorrector([("Network", 1), ("Available", 1)])
patterns.correct("The Netwxrk is Avxilable") # "The Network is Available"| Class | Description |
|---|---|
SpellCorrector |
Levenshtein-based correction (bundled dictionary or custom words) |
OCRCorrector |
Fix common OCR character confusions |
PatternCorrector |
Hyperscan-based multi-pattern matching |
Replacer |
Direct string replacement |
| Function | Description |
|---|---|
process_directory() |
Batch process files in parallel |
load_patterns() |
Load word,max_errors CSV |
load_replacements() |
Load search,replace CSV |
search-and-replace ./input -o ./output --patterns wordlist.csv
search-and-replace ./input --patterns patterns.csv --replacements replacements.csv -v -j 8| OCR Error | Fixed |
|---|---|
| 0 | O |
| 1, l, I | l |
| rn | m |
| cl | d |
| vv | w |
MIT