At a glance
OCR adds machine-recognized text; it does not establish that the recognition is correct. Compare the recovered layer with the page image, prioritize consequential text and record the exact inspected pages. Searchable or selectable text alone does not prove a usable reading order, correct tables or accessible forms.
Choose checks because of their meaning, not just whether the word appears in a dictionary.
| Image text | Possible recognition | Why to inspect |
|---|---|---|
| O / 0 | Letter or digit confused | Identifiers and account codes |
| 1.5 | 15 | Decimal changes magnitude |
| Not eligible | Eligible | Lost qualifier reverses meaning |
| Handwritten amendment | Omitted text | Important content may remain unread |
Scroll sideways to see all columns. Use arrow keys when the table is focused.
Describe which pages received OCR
A PDF can mix born-digital, scanned and previously OCR-processed pages. Determine which pages lack reliable text and which were actually processed. A sample of selectable text on the first page does not establish the state of the whole file.
Keep the source revision, OCR tool version and inspected page list. If a page failed recognition or was skipped, expose that state. Do not combine it with successful pages into an unexplained accuracy percentage.
- Identify image-only and mixed pages.
- Retain the original page images for comparison.
- Inspect recovered text on consequential and representative pages.
- Log corrections and unresolved recognition.
Check high-consequence content first
Compare dates, identifiers, decimal points, currency, units and words that qualify instructions. Use the page image at a useful zoom. Poor contrast, stamps, folds, skew and handwriting can make recognition less reliable, but a visually crisp scan can still contain errors.
Read complete phrases, not just isolated characters. OCR may capture every word while placing them in the wrong order, or join columns and table cells. Check whether a footnote belongs to the right statement and whether repeated headers interrupt the body.
- Numbers: verify signs, decimal marks, thousands separators and ranges.
- Instructions: compare not, unless, before and only.
- Names and identifiers: verify every consequential character.
- Tables: preserve relationships, not merely the cell text.
- Handwriting: retain uncertain or unavailable states and seek owner clarification.
Keep recognition review separate from structure review
Correct recognized text through a workflow that preserves evidence, then repeat the affected comparison. If the image is unreadable, ask for a better source instead of inventing missing wording. The QA log should identify what was observed and who resolved the ambiguity.
After text recovery, review headings, lists, reading order, language, table semantics and reader interaction. W3C’s OCR technique addresses obtaining text from scans; it is not a claim that the remaining accessibility work is complete. A searchable scan can still be difficult to navigate.
Sources and scope
Methods, worksheets and illustrative examples are DocAccessible editorial contributions. These references support the technical or product details described above. W3C techniques are informative guidance; they are not a certification scheme.
- W3C PDF7: OCR for scanned documents
Recovering actual text from scanned pages and checking the result.
- pypdf: text extraction
Why text extraction differs from OCR and cannot establish all visual or semantic meaning.
- W3C PDF3: reading and tab order
The relationship between PDF tag order, reading sequence and interactive elements.
Written by the DocAccessible team under our editorial policy. Suggest a correction if a source or product detail needs updating.