tesseract input.png outputbase

Tesseract’s docs still describe that as the simplest invocation; English is the default language, default output is plain text, and tesseract --help shows the options supported by your installed version. citeturn247668search0turn247668search1turn932099search4

For quick one-off screen captures, the most useful pattern is usually “read one image, print text to stdout”:

tesseract /tmp/cap.png stdout -l eng --psm 6

-l eng forces English, and choosing the right page segmentation mode matters a lot. In practice, --psm 6 is good for a uniform block of text, --psm 7 for a single text line, and --psm 8 for a single word. citeturn247668search0turn932099search2

If you want to restrict what Tesseract is allowed to recognize, use a whitelist with -c tessedit_char_whitelist=.... The official docs still recommend this when you know only a subset of characters should appear. citeturn932099search0turn932099search4

For magazine-style English, a practical whitelist is:

tesseract /tmp/cap.png stdout -l eng --psm 6 \
  -c tessedit_char_whitelist="ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789.,;:!?\"'()-"

That helps by ruling out math symbols and other junk, though it can also hurt if the real text contains characters you forgot to allow. That tradeoff is an inference from how whitelist filtering works. citeturn932099search0turn932099search4

A couple of generic examples:

# OCR a short paragraph clipping from a magazine column
tesseract /tmp/clip.png stdout -l eng --psm 6 \
  -c tessedit_char_whitelist="ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789.,;:!?\"'()-"

# OCR a single headline line
tesseract /tmp/headline.png stdout -l eng --psm 7 \
  -c tessedit_char_whitelist="ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789.,;:!?\"'()-"

# OCR a single big word, like a title fragment
tesseract /tmp/word.png stdout -l eng --psm 8 \
  -c tessedit_char_whitelist="ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz"

If results are messy, the two fastest fixes are usually: crop tighter, and switch the --psm to match what the image actually is. Tesseract’s own quality docs emphasize segmentation choice as a major accuracy factor. citeturn932099search0turn932099search2

Third option / omission: If you start doing this a lot, the next worthwhile step is preprocessing the screenshot first, because odd magazine fonts and background texture often hurt OCR more than the actual text does. Tesseract’s docs explicitly frame image quality improvement as part of getting better results. citeturn932099search0

tesseract /tmp/toc.png stdout -l eng --psm 6 \
  -c tessedit_char_whitelist="ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789.,;:!?\"'()-&" \
| xclip -selection clipboard

xclip -selection clipboard sends stdin to the normal clipboard, which is the one you paste with Ctrl+V. citeturn771239search0turn771239search6turn771239search15

On Wayland, use wl-copy:

tesseract /tmp/toc.png stdout -l eng --psm 6 \
  -c tessedit_char_whitelist="ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789.,;:!?\"'()-&" \
| wl-copy

wl-copy reads from stdin and puts that data on the Wayland clipboard. citeturn771239search1turn771239search13

For your use case, the practical defaults are:

• --psm 6 for a cropped block of table-of-contents text
• --psm 7 if the screencap is basically one line
• a whitelist like the one above to reduce garbage characters from stylized magazine-ish fonts. Tesseract’s command-line docs still recommend matching page segmentation to the layout, and whitelist settings are a standard quality-control trick when the character set is known in advance. citeturn463579search2turn463579search5

Two likely examples:

# block of contents entries
tesseract /tmp/toc-block.png stdout -l eng --psm 6 \
  -c tessedit_char_whitelist="ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789.,;:!?\"'()-&" \
| xclip -selection clipboard

# single entry line
tesseract /tmp/toc-line.png stdout -l eng --psm 7 \
  -c tessedit_char_whitelist="ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789.,;:!?\"'()-&" \
| wl-copy

Then just click your library program and hit Ctrl+V. citeturn771239search0turn771239search1turn463579search2

Third option / omission: If you want one command that works in either environment, make a tiny shell alias or script that tries wl-copy first and falls back to xclip. That is usually the least annoying long-term setup, because Wayland and X11 use different clipboard tools. citeturn771239search1turn771239search15

[Attachment reference unresolved: file-to]

tesseract /tmp/toc.png stdout -l eng --psm 6 \
| tr '[:upper:]' '[:lower:]' \
| wl-copy

Uppercase version:

... | tr '[:lower:]' '[:upper:]' | wl-copy

That’s the cleanest “no surprises” option.

Title case (slightly trickier)

There’s no perfect universal one-liner, but you’ve got a few workable pipeline options.

1. awk (good balance of simple + readable)

tesseract /tmp/toc.png stdout -l eng --psm 6 \
| awk '{for(i=1;i<=NF;i++) $i=toupper(substr($i,1,1)) tolower(substr($i,2)); print}' \
| wl-copy

This:

• splits on whitespace
• capitalizes first letter of each word
• lowercases the rest

It’s blunt (e.g., “of”, “the” still capitalized), but solid for TOC cleanup.

2. sed (lighter, slightly less robust)

... | sed -E 's/\b([a-z])/\U\1/g' | wl-copy

This capitalizes first letters but doesn’t force the rest lowercase, so combine if needed:

... | tr '[:upper:]' '[:lower:]' \
| sed -E 's/\b([a-z])/\U\1/g' \
| wl-copy

3. Python (more control, still pipeline-friendly)

... | python3 -c "import sys; print(sys.stdin.read().title())" | wl-copy

.title() is decent, but:

• turns “NASA” → “Nasa”
• same issue with “McDonald” etc.

Still often “good enough” for TOC cleanup.

Practical takeaway (your use case)

For OCR’d table-of-contents text:

• Lowercase normalization first
• Then optional title-case pass

Example “safe default”:

tesseract /tmp/toc.png stdout -l eng --psm 6 \
| tr '[:upper:]' '[:lower:]' \
| awk '{for(i=1;i<=NF;i++) $i=toupper(substr($i,1,1)) tolower(substr($i,2)); print}' \
| wl-copy

That gives you predictable formatting without weird OCR capitalization bleeding through.

Third option / omission

If you later care about proper title case (skip “of”, “and”, etc.), none of the above will do it cleanly. At that point, you’re into either:

• a custom Python script, or
• a dedicated text-processing tool/library

—but for TOC scraping, that’s usually overkill.