# Clean up OCR text Clean OCR text by expanding ligatures, rejoining hyphenated line breaks, reflowing lines, fixing confusables, and normalizing spacing. ## Run it - **CLI:** `gizza tool ocr-text-cleaner "The final report says HeIIo ,world .Next"` - **Web:** https://gizza.ai/tools/ocr-text-cleaner/ - **Agents:** machine-readable descriptor (parameters JSON Schema) at https://gizza.ai/tools/ocr-text-cleaner/tool.json ## Inputs - `text` — OCR text _(field)_ - `fix_ligatures` — Expand ligatures (fi → fi) _(field)_ - `join_hyphenated` — Rejoin words hyphenated across line breaks _(field)_ - `line_breaks` — Line breaks _(field)_ - `fix_confusables` — Fix letter/number confusables (HeIIo → Hello, l00 → 100) _(field)_ - `fix_rn` — Undo rn → m merge error (aggressive) _(field)_ - `fix_spacing` — Normalize spacing around punctuation _(field)_ ## Output - Cleaned text (text) ## Query parameters Open the tool pre-filled and auto-run via URL: - `text` — OCR text - `fix_ligatures` — Expand ligatures (fi → fi) - `join_hyphenated` — Rejoin words hyphenated across line breaks - `line_breaks` — Line breaks - `fix_confusables` — Fix letter/number confusables (HeIIo → Hello, l00 → 100) - `fix_rn` — Undo rn → m merge error (aggressive) - `fix_spacing` — Normalize spacing around punctuation Example: `https://gizza.ai/tools/ocr-text-cleaner/?text=The%20%EF%AC%81nal%20%20report%20says%20HeIIo%20%2Cworld%20.Next&fix_ligatures=true&join_hyphenated=true&line_breaks=keep&fix_confusables=true&fix_rn=true&fix_spacing=true` --- ## About this tool Optical character recognition is fast, but it is not clean. Text pulled from a scan, photo, or PDF arrives with a predictable set of **mechanical** errors: typographic ligatures that never got decomposed (`fi`, `fl`), words split by a hyphen at the end of a line, capital-I and lowercase-l standing in for each other, zeros posing as the letter O, columns wrapped mid-sentence, and stray or missing spaces around punctuation. This cleaner fixes exactly those errors — deterministically. There is **no dictionary and no language model**: every transformation is a fixed rule, so the same input always produces the same output and nothing is silently rewritten based on a guess about meaning. Toggle each fix on or off, choose how line breaks are handled, and the cleaned text updates live. Everything runs locally in your browser; your text is never uploaded. It pairs well with any OCR engine — run recognition first, then paste the raw output here to strip the noise before you proofread. ## Worked example Raw OCR output, with a ligature, a confusable-mangled word, a doubled space, and spaces on the wrong side of punctuation: ``` The final report says HeIIo ,world .Next ``` With **Expand ligatures**, **Fix confusables**, and **Normalize spacing** on (the defaults), the tool returns: ``` The final report says Hello, world. Next ``` `fi` became `fi`, `HeIIo` (capital-I confusables) became `Hello`, the double space collapsed, the space before each punctuation mark moved to after it, and a missing sentence space was inserted before `Next`. ## Line-break handling Scanned columns often wrap every physical line, which is wrong once the text is reflowed. The **Line breaks** control offers three modes: - **Keep as-is** — only line-ending characters are normalized (`\r\n` → `\n`). - **Reflow within paragraphs** — single line breaks inside a paragraph become spaces, while blank-line paragraph breaks are preserved. This is the usual "un-wrap a scanned column" fix. - **Collapse to one line** — every line break becomes a single space. Words hyphenated across a line break (`para-` / `graph`) are rejoined into `paragraph` when **Rejoin words hyphenated across line breaks** is on. A hyphen followed by an uppercase word is treated as a real compound (`New-` / `York` → `New-York`), so the hyphen is kept. ## Limits & edge cases - **Input size:** up to 200,000 characters per run. Larger documents should be split into chunks. - **Confusables are context-based, not dictionary-based.** A digit is only treated as an in-word letter when it sits *between* two letters, which protects alphanumeric codes like `COVID19`, `MP3`, and `B12`. The pronoun `I`, capitalized sentence starts, and all-caps acronyms are left alone. - **`rn → m` is aggressive and off by default.** With no dictionary it cannot tell `rn`-in-`modern` from `rn`-that-should-be-`m`, so enabling it will also rewrite real words. Use it only on text you know is affected. - **Non-Latin text is preserved.** Accents, CJK, and emoji pass through untouched; the tool only rewrites the specific Latin confusables and ligatures it knows about. - **Curly-quote / smart-quote straightening is intentionally not included** — that is a separate concern from OCR mechanical errors. ## FAQ
Does this tool spell-check or correct real words? No. It only fixes *mechanical* OCR errors with fixed rules — ligatures, hyphenation, line breaks, character confusables, and spacing. It has no dictionary, so it will not insert a missing space between two glued real words (`thejob`), add a missing apostrophe (`dont`), or correct a genuine misspelling. That keeps every change predictable and reversible in your head.
Why did it leave "COVID19" and "MP3" alone? Confusable fixing is context-aware. A digit is only converted to a letter when it is surrounded by letters on both sides, and a letter is only converted to a digit inside a number-dominated token. `COVID19` and `MP3` end in digits that are not flanked by letters, so they are recognized as codes and left untouched — while `l00` (letters posing as digits) still becomes `100`.
What is the "rn → m" option and why is it off by default? `rn` is one of the most common OCR merge errors: the letter `m` gets misread as the two letters `r` and `n`, turning `modem` into `rnodem`. Undoing it is valuable, but with no dictionary the fix is blind — it replaces *every* `rn`, including the ones inside legitimate words like `modern` or `turn`. Because that can corrupt good text, it ships off; turn it on only for text you know suffers from the merge.
Is my text uploaded anywhere? No. The cleaner is compiled to WebAssembly and runs entirely in your browser tab. Nothing you paste leaves your device, which makes it safe for confidential scans and works offline once the page has loaded.
## Related tools - [Text Wrap](https://gizza.ai/tools/text-wrap/): Hard-wrap text to a fixed column width in your browser — keep indentation and blank lines, optionally break long words. Free, private, no sign-up. - [Unwrap text](https://gizza.ai/tools/unwrap-text/): Remove hard line breaks inside paragraphs to rejoin wrapped text into continuous lines, in your browser. Nothing is uploaded. - [Extract Action Items from Meeting Notes](https://gizza.ai/tools/action-item-extractor/): Extract action items, owners, and decisions from meeting notes or daily notes with deterministic rules. Markdown checklist or JSON, private in-browser. - [Add Line Numbers](https://gizza.ai/tools/add-line-numbers/): Add line numbers to every line of text online, like nl or cat -n — custom start, step, separator, and alignment. Free and private, runs in your browser. - [ANSI Log Renderer](https://gizza.ai/tools/ansi-log-renderer/): Paste ANSI-colored terminal output or CI logs and render them as HTML, or strip escape codes to plain text. Handles 16-color, 256-color, and truecolor SGR codes.