Run OCR so scanned PDFs contain real text
A scanned PDF is a stack of photographs of pages. There is no text in it until somebody runs optical character recognition. Until then, as W3C puts it, assistive technologies cannot read or extract the words, and nobody can select, edit, resize, or reflow the text or change its colors. W3C publishes PDF7 as a sufficient technique, so running it satisfies the rule and it is one route rather than the only one. Acrobat Pro's Scan and OCR tools recognize the text and flag what it calls OCR suspects for you to correct one at a time, and running a file through OCR may add tags too. Rebuilding from the source document is better wherever the source still exists. Scans are everywhere in legal and government work. Signed contracts, court filings, archived records.
How we find it in an audit
The screen reader test takes seconds. We open the PDF and try to read, and an image-only scan produces silence or "blank page". Failing to select any text and finding an empty tag tree confirm it. Then we save the document as text and read what comes out, which shows both the recognition quality and the order, and is the fastest way to catch OCR that looks fine on screen and is garbage underneath.
How affected users experience it
There is no more complete failure in PDF accessibility than an image-only file. The screen reader opens a 30-page signed agreement and reads nothing at all. Not scrambled, not partial. Silence, or "empty document". The person holds a legally binding document they cannot read one word of.
Passes vs. fails
Passes
The scan is run through Acrobat Pro's Recognize Text, the OCR suspects are corrected against the original, and the result is tagged. Better still, the original Word file is exported as a tagged PDF with only the signature page appended as a described image.
Fails
A signed settlement agreement is scanned to PDF and emailed to all parties. Screen readers report every page as blank.
How this gets tested
The W3C publishes test rules that define what a checker looks for here.
- HTML images contain no textNothing implements this yet
Other ways to satisfy this rule
9 guides on this site are filed under 1.4.5 Images of Text. W3C lists this one as sufficient for that rule on its own. Implement it correctly, in a way your readers' software actually supports, and the rule is met.
- C22sufficientStyle real text instead of text images
- C30sufficientLet users swap images of text for text
- G140sufficientKeep content and structure separate from presentation
- C6advisoryKeep source order meaningful when CSS repositions content
- C8advisorySpace letters with CSS, not blank characters
- C12advisorySize text in percent so it scales
This guide is our interpretation of W3C technique PDF7: Performing OCR on a scanned PDF document to provide actual text. W3C publishes its techniques as guidance rather than as the standard, and says so on every one of them. The success criterion is what conformance is measured against, and a technique is one documented way to meet it.