Taggart rule TG-SCAN-001

Scanned PDF page with no text layer: a picture of text

A scanned page is a photograph of a page. There is no text object in the file at all, so there is nothing to tag, nothing to put in reading order and nothing to read aloud — every structural repair in this engine is a no-op on it, and a report listing thirty tagging failures describes the symptom and misses the disease. It is also the commonest thing in a real remediation backlog and the repair with the largest single gain, because the document goes from unreadable to readable rather than from badly tagged to well tagged. Recognition writes the words underneath the picture as an invisible layer; the image itself is not touched, so the page still renders exactly as it did.

What a screen reader actually does

The reader announces "graphic" and stops, or says nothing at all. Nothing on the page can be searched, selected or copied either — for anyone using assistive technology the document simply is not there.

The cause, not the symptoms

A scanned page fails a long list of checks — untagged content, an untagged image, no reading order — and every one of them follows from a single fact: there is no text. The protocol’s OCR conditions, 08-001 for recognition errors and 08-002 for untagged recognised text, only apply once there is text to judge. This finding names the cause, so a report on a scan leads with it rather than with thirty tagging failures.

What is in the file

A scanner, or a “print to PDF” of an image, produces a page whose content stream draws one picture:

q 595 0 0 842 0 0 cm /Im0 Do Q          % one full-page image, and nothing else

There is no BT … ET, no font, no text object. The words exist only as pixels. To a screen reader the page is a single graphic — announced as one if it is tagged, and not announced at all if it is not. Search finds nothing, selection selects nothing, and a reader that wants to enlarge or reflow the text has no text to work with.

How to check it free

Search for a word you can see. Ctrl/Cmd + F. No match on a word that is plainly on the page means there is no text there.

Try to select a line. On a scan the selection is a rectangle of image, or nothing.

Acrobat Pro’s Accessibility Checker. Adobe documents an Image-only PDF rule that reports whether the document contains non-text content that is not accessible, and notes that a document which appears to contain text but has no fonts may be an image-only PDF (Adobe: Create and verify PDF accessibility). Adobe also notes that if you try to select text or use Read Out Loud on a page that has not been through recognition, Acrobat offers to run it.

pdffonts (Poppler, free). On a scanned document it lists no fonts at all.

How to fix it free, in Acrobat Pro

Adobe documents recognition as (Adobe: Recognize text in scanned documents):

1. Open the scanned PDF and select All tools → Scan & OCR.
2. Select In this file.
3. Choose the page range and the language of the text.
4. Optionally, open Settings for further options.
5. Select Recognize Text.

Choose the language deliberately. Recognition is guided by it, and a Hebrew or Arabic page read as English produces Latin-letter guesses.

Then check what it produced. Adobe documents All tools → Scan & OCR → Correct recognized text, ticking Review recognized text, which steps through the words Acrobat was unsure of and lets you correct each one in a Recognized as field before selecting Accept (Adobe: Fix text issues in scanned PDFs).

Recognition gives you text, not structure. The result still has to be tagged — Adobe documents All tools → Prepare for accessibility → Automatically tag PDF for a first pass — and the reading order checked, exactly as for any untagged document. Tagging an untagged PDF and remediating a scanned PDF cover what comes after.

If the paper original is to hand and the scan is poor, rescanning at a higher resolution often does more for accuracy than any amount of correction afterwards.

What Taggart does

The check. A page is reported when it paints anything — an image, a path, any drawing — and holds fewer than 24 extractable characters:

Page 1 paints content but holds no extractable text, so it is an image of a page rather than a page.

Twenty-four is a threshold, not a definition: a scanner’s own header stamp is two or three characters, and a scanned page should not escape because of one. The cost is in the other direction. A real page with very little text on it — a cover with a short title, a section divider, a page that is one chart — is reported too, and the message says “no extractable text” even when there are a few characters. The finding’s evidence gives the actual count; read it before acting.

The repair, ocr-text-layer, is proposed and waits for approval rather than running in Fix All Safe. When approved:

• It runs Tesseract on each page the check reported, rendered at 300 dpi, reading English, Hebrew and Arabic together — whichever of the three are installed on the server.
• It writes the recognised words as invisible text — text render mode 3 — over the image, in the page’s own coordinate space. The image is not touched, re-encoded or re-drawn, so the page renders as it did.
• It skips a page that already holds 24 or more characters, because recognising a page that has text lays a second copy of every word under the first and a screen reader reads it twice.
• In the same pass it tags the new words, and marks the full-page image as a background artifact, because the text over it is what it shows. Headings and paragraphs are proposed from the size and position of the text, and the change log points you to the reading-order editor to check them.

Two things it leaves to a person. Recognition errors are 08-001, and no tool can find a wrong word that looks plausible — read the result against the page. And recognition runs on every page the check reported, with no way to leave one out. On a page reported only because its text is short, it reads the words that are already there and lays an invisible duplicate with them, so the title is read twice. If the evidence on any reported page shows real characters, check that page in the output and expect to remove the copy.

If recognition is not installed on the server, the repair says so and changes nothing.

How Taggart handles it

Taggart checks for this condition automatically and reports it as a blocker, with where in the file it found it.

The fix is Read the scanned pages and add a text layer. It changes what the document means, so Taggart proposes it and waits for your approval rather than applying it unattended.

Where this sits in the standards

Related conditions

• 08-001 OCR-generated text contains significant errorsThe invisible text layer produced by OCR does not match the visible glyphs. Confusions like rn→m, l→1, and dropped Hebrew niqqud are typical.
• 08-002 OCR-generated text is not taggedOCR added a text layer but no structure tree, so the document is searchable but not navigable.
• 01-005 Content is neither marked as Artifact nor tagged as real contentEvery text-showing, image-painting and path-painting operator on a page must sit either inside a marked-content sequence that maps to the structure tree, or inside an artifact sequence. Content that is in neither is orphaned.
• 13-001 Graphics objects are not tagged with a Figure tagAn image or vector drawing that carries meaning is not inside a Figure element. Either it is untagged, or it has been swept into a paragraph.
• 31-027 A font dictionary does not contain the ToUnicode entry and no exemption appliesA ToUnicode CMap is required unless the font uses MacRoman/MacExpert/WinAnsi encoding, or is a Type 1/Type 3 whose glyph names are all in the Adobe Glyph List or the Symbol set, or is a Type 0 whose descendant uses a standard Adobe character collection, or is a non-symbolic TrueType.

Other conditions in checkpoint TG-SCAN

Checkpoint TG-SCAN covers scanned documents. These are its other failure conditions; the ones without a link are in the rule set but do not have a written page yet.

Checked with Taggart on a generated file that fails this condition on 2026-10-02. The Acrobat steps follow Adobe’s documentation and were not walked through by hand. Last updated 2026-10-02.