tesseract

• tesseract-ocr/tesseract
  • Apache-2.0, C++
  • Tesseract 主要是为印刷体文本设计的, 有提供手写体识别
  • 2017 4.0 引入 LSTM
  • 2021 5.0 代码优化、指令集加速
  • tessdata 由 Google 维护训练
• 目前不是最准确的 而是 最通用、最轻量、最容易嵌入

# tesseract-lang -> https://github.com/tesseract-ocr/tessdata_fast/
brew install tesseract

tesseract img.png stdout

# /opt/homebrew/share/tessdata/
# -l eng+chi_sim
tesseract input.jpg -l chi_sim output hocr

• tesseract-ocr/tessdata_fast
  • Fast integer versions of trained LSTM models
• tesseract-ocr/tessdata_best
  • Best (most accurate) trained LSTM models.
• tesseract-ocr/tessdata
  • Trained models with fast variant of the "best" LSTM models + legacy models
• tesseract-ocr/langdata
• Static Binary
  • DanielMYT/tesseract-static
    • Linux 大约 10MB

• 输出格式支持
  • text, hOCR (HTML), PDF, invisible-text-only PDF, TSV, ALTO , PAGE
  • ALTO - Analyzed Layout and Text Object XML
• ./
  • configs
  • eng.traineddata
  • osd.traineddata - Orientation and Script Detection
  • pdf.ttf
    • GlyphLessFont
  • tessconfigs
  • snum.traineddata

tesseract.js

• naptha/tesseract.js
  • Apache-2.0
  • npm tesseract.js
  • worker+lib 约 200Kb
  • 模型 osd.traineddata 约 10MB
• naptha/tesseract.js-core
  • wasm 引擎