Prepare a scanned PDF for OCR and check the recognized text

Decide what the output must do

Before scanning or running OCR, decide whether you need a searchable archive, editable text, data from a table, or a document that supports an accessibility requirement. These goals need different checks. Searching for a name is a smaller task than relying on every number in a long invoice or producing a fully accessible document.

Keep the original scan and save recognized output separately. If you will process multiple documents, use a simple naming convention and record which files have been reviewed. A filename such as invoice-ocr-unchecked.pdf is more useful than assuming every completed processing job is ready to distribute.

Improve the source before recognition

Inspect every page at a readable size. Look for missing edges, shadows, glare, a curved page near a book’s spine, and text lost in a dark background. If the source is still available, recapture an unreadable page rather than expecting OCR to reconstruct characters that cannot be seen.

For a camera capture, position the page flat, keep the camera square to it, and use even lighting. Avoid fingers, staples, or other objects covering text. For a scanner, test one representative page before scanning a whole batch. Use enough detail to distinguish the smallest meaningful text without creating unnecessarily large images.

Rotate sideways pages and correct visible skew where your tool supports it. Crop excess background carefully, leaving all page content intact. Check that image enhancement has not erased faint pencil marks, light punctuation, or a pale stamp. Keep color when it conveys information, such as annotations or highlighted categories.

Choose recognition settings deliberately

Select the language or languages actually present, where supported, and confirm the intended page range. A document may contain a typed cover, scanned attachments, and pages with an existing text layer. Check the tool’s behavior for mixed files rather than assuming the entire document needs the same treatment.

Choose an output that matches the task. A searchable PDF can preserve page images while adding recognized text; an editable export may reconstruct the layout. These are different deliverables. For important records, retain a visual reference so that anyone using extracted text can check the source.

Do a small trial on a difficult page first. Compare the recognized text before committing time to a long document. If results are poor, improve the capture or settings and rerun that sample. Repeating the same process on the same blurred image rarely addresses the underlying problem.

Review more than the highlighted errors

Adobe documents an OCR review workflow that flags uncertain words and lets you compare the recognized text with the image. Use that feature when available, but do not limit review to flagged words. A confidently recognized character can still be wrong.

Search for a known phrase on several pages, then copy representative passages into a plain-text editor. Compare the extracted text with the page image. Include the smallest type, text near page edges, a table, and any page whose scan quality is noticeably worse.

Pay special attention to characters that can be confused: letter O and zero, lowercase l and the number one, punctuation, currency symbols, and decimal points. Verify names, dates, identifiers, amounts, and units directly against the scan. When the information will support a consequential decision, review all relevant fields rather than relying on a spot check.

For columns and tables, check the order of extracted content. Words can be recognized correctly while rows, labels, or columns are combined incorrectly. Handwriting, formulas, and unusual fonts deserve additional scrutiny. If a section remains unclear, mark it for manual review instead of silently guessing.

Keep OCR separate from accessibility and archival validation

The Section508.gov training material describes OCR correction as part of a larger process that also includes tagging, image alternatives, and conformance testing. Searchable text alone does not make a PDF accessible. Reading order, headings, tables, and other structure require their own checks.

Likewise, do not label a document PDF/A-compliant merely because OCR completed. If an archive or recipient requires a particular standard, use the appropriate creation and validation workflow for that requirement. State what you actually verified rather than claiming blanket compliance.

Finish with a usable, reviewed copy

Open the saved output after processing. Confirm page count, orientation, search behavior, and the fields you corrected. If you compress or reorganize it afterward, check the resulting file again because later processing can change or remove the text layer.

Keep the original scan, the reviewed output, and any correction notes together in your own document-management system. When sharing extracted text, identify remaining uncertainty clearly. A smaller, accurately reviewed extract is often more useful than a large block of unchecked OCR text.