SCANNED PDFS
Convert scanned PDF tables to Excel with OCR
When a price list is a photograph of a page, there is no real text to copy. Linework can rasterise the page and read it with English OCR in your browser, then you review the rows.
When OCR is used
Automatic mode runs OCR on pages with fewer than about 50 non-whitespace text characters. That catches image-only pages while leaving text pages on the fast path. If a page has selectable headings but the products are a scan, automatic mode may miss the table. In that case choose “OCR on every page” and use Re-read PDF.
OCR is lazy-loaded. The Tesseract worker and English language data download from this site only when a page needs them. They run locally. Document contents are not sent to an external recognition API.
What works best
- Clear scans with good contrast
- Upright pages, not rotated or skewed
- English text
- A real table: code or description and a unit price
Skewed phone photos, handwriting, stamps over prices and very small type all increase errors. Linework does not deskew or enhance a poor scan beyond rendering the PDF page.
What OCR gets wrong
OCR can confuse characters. Product codes are high risk: 0 and O, 1 and I, 5 and S, 8 and B. Prices are high risk: commas, points and extra digits. Pack sizes such as “6 x 1 L” can split across lines. Rows with any word confidence below 85 are flagged for closer review. That flag is not calibrated accuracy. Treat every OCR row as unchecked until you compare it with the page.
Limits
At most 20 pages in one file are processed with OCR. The file must also stay within 50 pages and 15 MB. If a scan is longer, split it. Candidate rows are capped at 3,000. Cancel stops a long recognition run and keeps the previously loaded document.
After recognition
The rest of the workflow matches text PDFs: pick the header, map columns, set decimal format, edit, merge wrapped descriptions, mark rows checked, download Excel. See how Linework works, PDF table to Excel, and supplier price lists.
Time, cancellation and re-reading
OCR is slower than text extraction because each page is rendered and recognised in sequence. A few pages may finish in under a minute; twenty pages can take several minutes on a modest machine. Progress is shown while the worker runs. Cancel stops the current run and keeps the document you already had. A failed or cancelled scan does not overwrite a good extract.
If the first pass missed products, change the reading mode and use Re-read PDF. Re-reading replaces the extract after you confirm, including edits, so export first if you already corrected rows you need.
When not to use OCR
Do not OCR a digital text PDF “just in case”. Recognition will introduce errors that the original text did not have. Use selectable text only when the table is already text. Use OCR when the products are an image, or when automatic mode left pages empty. Handwritten annotations, stamps over prices and severe skew are better fixed by requesting a digital list than by forcing the scan through another pass.
English is the only OCR language here. A Greek, German or mixed-language scan may return leftover Latin lookalikes. That is not a translation feature and should not be treated as one.
Open a scanned English price list and choose OCR on every page if products are missing.
Frequently asked questions
When does Linework use OCR?
Automatic mode uses OCR on pages with very little selectable text. You can also choose OCR on every page and re-read the PDF.
Which languages are supported?
OCR is English-only in this version. Non-English scans are outside the current scope.
How accurate is OCR?
Accuracy depends on scan quality. Characters can be confused. Confidence flags are engine signals, not a guarantee. Every row still requires review.
What are the OCR limits?
Up to 20 pages can be processed with OCR in one file, within the overall 50-page and 15 MB limits.
Use automatic mode, or OCR on every page for English scans.