SHAPECLUSTERING(1)

🌸 NAME — shapeclustering(1) 🌸

okay so shapeclustering is literally shape clustering training for Tesseract. That's the whole NAME vibe, bestie. 💖

---

✨ SYNOPSIS ✨

shapeclustering -D output_dir -U unicharset -O mfunicharset -F
font_props -X xheights FILE...

🦋 DESCRIPTION 🦋

shapeclustering(1) takes extracted feature .tr files — those are generated by tesseract(1) run in a special mode from box files — and produces a file shapetable plus an enhanced unicharset. The way it eats: feature files in, iconic output files out. ✨

Vibe check though: this program is still experimental, and it is not required (yet) for training Tesseract. No cap, that's straight from the manual. 💅

💅 OPTIONS 💅

• -U FILE — the unicharset generated by unicharset_extractor(1). That's the one, bestie. 🫶
• -D dir — directory to write output files to. Yes, dir, the place where the output files go. 📁
• -F font_properties_file — (Input) font properties file, where each line is of this form, and each field other than the font name is 0 or 1:

'font_name' 'italic' 'bold' 'fixed_pitch' 'serif' 'fraktur'

• -X xheights_file — (Input) x heights file, each line is of this form, where xheight is calculated as the pixel x height of a character drawn at 32pt on 300 dpi. [ That is, if base x height + ascenders + descenders = 133, how much is x height? ] Like, do the math, bestie. 🧮

'font_name' 'xheight'

• -O FILE — the output unicharset that will be given to combine_tessdata(1). Periodt. 👑

🔥 SEE ALSO 🔥

• tesseract(1)
• cntraining(1)
• unicharset_extractor(1)
• combine_tessdata(1)
• unicharset(5)
• https://tesseract-ocr.github.io/tessdoc/Training-Tesseract.html

🎀 COPYING 🎀

Copyright (C) Google, 2011 Licensed under the Apache License, Version 2.0 — yes bestie, that's the legal tea. ⚖️

👑 AUTHOR 👑

The Tesseract OCR engine was written by Ray Smith and his research groups at Hewlett Packard (1985-1995) and Google (2006-2018). Absolute legend behavior. 🌈

---

xoxo, bestie — page date 2026-09-28, manual SHAPECLUSTERING(1) 💖🎀