# Can OCR Read Strikethrough Text in a Scanned Redline?

- Author: Schema Law (https://www.schemalaw.com/)
- Published: 2024-10-03T10:23:35Z
- Updated: 2026-10-11T19:22:20Z
- Category: Difficult scans
- Canonical: https://www.schemalaw.com/blog/can-ocr-read-strikethrough-text

> Can OCR read strikethrough text? It reads the struck words with errors and may drop the line, so deletions look live. See what to check in a scanned redline.

OCR can read strikethrough text, in the sense that it returns the words under the line, with some letters wrong. What it may not return is the line. Converted text can present a struck sentence as ordinary text, so a deletion in a scanned redline reads as if it were still part of the document.

## Can OCR read strikethrough text?

Partly. Printed strikethrough is a thin rule through the middle of each letter, and the letters stay visible above and below it. Software recovers some of the words and misreads others. Whether the output records that the words were struck is a separate matter, and it is the one that changes meaning.

So there are two failures to look for, and they pull in opposite directions:

- **The struck words are garbled.** You see nonsense in the output and know to look at the page.
- **The struck words are read perfectly and the strike is lost.** The output looks clean. A deleted clause sits in the text as if it were agreed.

The second is worse because nothing looks wrong. Schema Law's site says that nothing marks a passage as uncertain, and you should assume the same of any converter until you have tested it. Read a redline's converted text with the page image open beside it.

This post covers printed and scanned redlines. Pen strikes on a signed page are a different job, covered in the guide to [handwritten changes to a contract](https://www.schemalaw.com/blog/handwritten-changes-to-a-contract).

## What does a line through a word do to recognition?

The line joins every letter in the word at mid-height. Recognition separates letters by the white space between them and identifies each by its strokes. A strike removes the gaps and adds a horizontal stroke to every character, so letters merge and several turn into other letters.

The misreads follow from where the line falls:

| Letter on the page, with a line through it | What it can become | Why |
| --- | --- | --- |
| c, o | e | The line supplies the bar of an e |
| l, I, 1 | t, f, + | A crossbar appears on a plain stem |
| n, h | A or H | The bar links two stems |
| space between words | a hyphen or dash | The line shows alone in the gap |
| period, comma | lost | The mark is close to the line and merges into it |
| 3, 5, 6 | 8, or a letter | The line closes an open curve |
| whole word | one long token, or two halves | Letter gaps are gone |

A fictional clause shows the range. The page reads "within thirty (30) sixty (60) days", with a printed line through "thirty (30)". Three converters, or one converter on three pages, might return:

```text
within thirty (30) sixty (60) days
within thirt-y-(80) sixty (60) days
within sixty (60) days
```

The first keeps both numbers and no sign of which is live. The second signals trouble. The third dropped the struck words, which reads correctly here and tells you nothing about what was deleted.

On a proposed order or an amended complaint, the answer's struck sentence has to stay visible as struck. A counterclaim lined out in the margin, or a motion to dismiss with a paragraph ruled through, still has to show the deleted words. Type them from the page with a struck label, so a reader of the Word file cannot treat the deletion as live text.

Weight and position matter. A line through the exact middle of small type does more harm than one that rides high. A strike that was photocopied until it thickened covers more of each letter. Doubled strike lines cover more again. Scan resolution matters too, since at low resolution a thin rule and the strokes beside it blur into one band; the post on [the best DPI for OCR](https://www.schemalaw.com/blog/best-dpi-for-ocr) covers how much small print needs.

## Does converted text keep the strikethrough and underline of a scanned redline?

Do not assume it does. Strikethrough and underline are formatting. Plain text, CSV and JSON have no way to hold them. Word, HTML and Markdown could hold them, but whether a converter detects the marks on a scan and writes them into the file varies by tool, so test one page of your own.

In a redline the formatting is the content. Microsoft's guide to [tracking changes in Word](https://support.microsoft.com/en-us/office/track-changes-in-word-197ba630-0f5f-4a8e-9a77-3712475e806a) describes the convention in that program: "Deletions are marked with a strikethrough, and additions are marked with an underline." Print that document and scan it, and the only thing separating old language from new is a line through or under the words. Remove the lines and the two versions run together in one paragraph.

What each kind of output can carry:

| Output | Can it hold a strike? | What to expect from a scanned redline |
| --- | --- | --- |
| Plain text | No | Old and new words in sequence |
| Word | Yes, as character formatting | Test it; do not rely on marks carrying over |
| Searchable PDF | The page picture shows it | Search finds struck words exactly as it finds live ones |
| Excel, CSV, JSON | No | Words only |

The searchable PDF row deserves a second look. The page image keeps every line, so a reader sees the redline as printed. The hidden text does not know about the lines. A search for "thirty" lands on the struck word, and copying the paragraph copies both versions. Schema Law describes that export as your original file with a hidden text layer, and the same behavior follows for any file built that way. The post on [whether OCR keeps formatting](https://www.schemalaw.com/blog/does-ocr-keep-formatting) covers which structure survives conversion in general.

### A one-page test

1. Pick a page with a struck phrase, an underlined insertion and an untouched paragraph.
2. Convert it and export Word and plain text.
3. In the Word file, check whether the struck phrase is present, and whether it is formatted as struck.
4. In the plain text, find where the struck phrase and its replacement sit relative to each other.
5. Write down what you found, so everyone on the matter reads the converted redlines the same way.

Schema Law's free plan takes one document a day of up to three pages, enough for this test. Every plan includes Word and plain text among its export formats, and the [pricing page](https://www.schemalaw.com/pricing) lists the page limits.

## Underlines that touch letters and change bars in a printed redline

Underlines and margin bars cause their own errors. An underline that touches the bottoms of letters clips the tails of g, j, p, q and y, or comes back as a row of underscore characters. A change bar in the margin is a vertical stroke, and software may read it as a letter and attach it to the line.

What to look for:

- **Clipped descenders.** With its tail merged into an underline, "y" can read as "v", "g" as "a" or "o", and "p" as "n". Check underlined names and defined terms.
- **Underscores in the text.** An underline read as characters gives "the_Tenant_shall", or a line of underscores on its own row beneath the words.
- **A comma or period swallowed by the underline.** Punctuation at the baseline is the first thing to go.
- **Underlines taken for table rules.** A run of underlined lines can be returned as rows of a table.
- **Change bars as characters.** A vertical bar in the margin can arrive as "|", "l", "I" or "1" at the start or end of a line. In a numbered list, a stray "1" in front of a paragraph number is easy to miss.
- **Margin notes pulled into the body.** Microsoft's guide says that revisions shown as balloons "display in the margins of the document". On a scan, that margin text may be woven into the paragraph lines beside it. The post on [reading order in two-column documents](https://www.schemalaw.com/blog/can-ocr-handle-two-column-documents) explains how side-by-side text gets interleaved.
- **Colored redlines scanned in black and white.** If insertions were blue and deletions red, a black and white scan makes them the same. The comparison of [grayscale and black and white scanning](https://www.schemalaw.com/blog/grayscale-vs-black-and-white-scanning) covers what color keeps.

Shading behind changed text is a related mark. See the post on [whether OCR can read highlighted text](https://www.schemalaw.com/blog/can-ocr-read-highlighted-text).

## Struck provisions in bills, orders and marked-up drafts

Struck text reaches a legal file in three main forms: legislative bills that show amendments to existing law, court orders and proposed orders with language lined out, and printed drafts exchanged in a negotiation. In each, record a deletion in the converted text explicitly, because the text alone will not.

### What to check in each

- **Bills and amendments.** Find the legend or drafting key that says what strikethrough, underline, brackets and italics mean in that legislature's documents. Conventions differ, so do not carry one body's key over to another.
- **Orders.** A judge or clerk may line out part of a proposed order. A ruled, printed strike and a pen strike can sit on the same page. Read the whole order from the image; the struck words are part of what the court did.
- **Marked-up drafts.** Look for a legend on the first or last page of a comparison printout. If you hold both versions as separate documents, comparing them yourself is more dependable than reading a scan of someone else's redline; see the guide on [how to compare two scanned contracts](https://www.schemalaw.com/blog/how-to-compare-two-scanned-contracts).

### Recording a deletion

In a working copy, mark struck words so that a reader of plain text cannot mistake them for live ones. The National Archives' [transcription tips](https://www.archives.gov/citizen-archivist/transcribe/tips) offer a simple form: "You may include [crossed out] next to crossed out text to indicate its format". For example:

```text
within [struck: thirty (30)] sixty (60) days
[struck paragraph 4(b), 6 lines; see page image]
```

Type the struck words from the image, not from the converted text. If a strike makes words unreadable, write "[struck, illegible]". The post on [marking in converted text what the page shows but the text cannot](https://www.schemalaw.com/blog/ocr-redacted-documents#marking-each-redaction-in-the-converted-text-of-a-document) sets out one convention for struck words, seals and redactions across a matter.

### Ask for the file behind the paper

If the redline was printed from a word processor, a file exists somewhere. Before you spend an hour repairing a scan, ask the sender for that file, or for clean copies of both versions. What you are entitled to, and in what form, depends on your agreement, your court's rules and the stage of the matter. A request costs one email, and the scan remains your record of what was sent.

## Try it on a real document

Upload a scanned PDF and export it as Word, a searchable PDF or plain text on the free plan. Your documents are never used to train an AI model. [Start free](https://www.schemalaw.com/api/auth/google)