OmniDocExtract Tables from Scanned PDF and Images with OCR
Learn how to pull a table out of a scanned PDF or image with OCR, so you can copy the cells or export them to Excel or Word, and how to check the result.
If you need to extract a table from an image or a scanned PDF, you have probably already tried to copy it and got nothing, or a jumble of stray characters. A scan stores the table as a picture. There is no text layer for a copy command to read, so the cells stay locked until someone recognizes the characters. This is the gap that optical character recognition (OCR) fills, and it is why a plain text extraction misses these tables.
This guide is for anyone who has a table in a scanned report, a photographed form, or a screenshot and needs the numbers in a spreadsheet or a Word file. You will see how to run OCR in the OmniDoc desktop client, how to handle image files, and how to check the output before you rely on it. The steps use the desktop PDF editor, which can OCR scanned PDFs and export to Excel and Word.
How do you extract a table from an image with OCR? The short answer
Run OCR on the page that holds the table, then export or copy the recognized cells. If you start with a picture file, convert it to PDF first, because OCR in the desktop client works on PDF pages.
| Your situation | What to do | Why |
|---|---|---|
| Scanned PDF where the table is a picture | Run OCR on the table pages, then export to Excel | OCR turns the picture into text the export can read |
| Photo or screenshot (PNG or JPG) | Save it as a PDF, then run OCR on that PDF | The OCR step works on PDF pages |
| PDF where you can already select the cells | Skip OCR and export or copy directly | The text layer already exists, so OCR adds little |
| Table is mostly prose or a form | Export to Word instead of Excel | Word keeps the text in document form |
| Table has merged headers or many columns | Export, then rebuild the structure in the target app | Recognition can flatten complex layouts |
Before you start
- Check whether the PDF is scanned. Try selecting one line of text. If nothing highlights, the page is an image and needs OCR. This matters because running OCR on a file that already has text adds a step without helping.
- Use the cleanest scan you have. OCR reads characters, so blur, skew, and shadows cause misread digits. A straight, evenly lit scan at a readable resolution gives OCR a good starting point.
- Convert images to PDF. Use your operating system's print-to-PDF option or any image-to-PDF tool. Then load the new PDF into the desktop client. Keep the file names clear so you can tell the source images apart later.
- Decide your output format. Excel suits tables of numbers you want to sort or total. Word suits tables inside a report you are editing as text.
- Keep the original file. You will compare the export against it, and you may need to re-run OCR on a better scan if something is wrong.
Running OCR on a scanned table, step by step
- Install the desktop client if you do not have it. Download it from download.omnidoc.top. It runs on Windows, macOS, and Linux, and includes the PDF editor with OCR.
- Open your PDF in the PDF editor. If your source was an image, open the PDF you created from it. Scanned PDFs and converted images behave the same way at this stage.
- Fix the page orientation if needed. If the table is sideways or upside down, rotate the page first. Recognition is more reliable when the rows run straight across the page.
- Run OCR on the pages with the table. You do not need to process the whole document. Limiting OCR to the table pages keeps the job small and makes the results easier to review.
- Find the recognized table. Select the table area and check that the cells now behave as text. If they do not, the page may be too blurry or skewed, and a better scan is the next step.
- Export to Excel or Word. Use the export option for the format you chose. Excel works well for numeric tables, and Word works well for tables you plan to edit as part of a document.
- Open the exported file and check it. Look at the rows and columns, the headers, and a few numbers you can verify against the original.
- Correct errors in the exported file. Fix misread digits and split or merge cells in Excel or Word. Editing the exported file is usually quicker than returning to the scan.
Check the result
OCR output can contain errors, especially with small type, faded print, or dense columns. Check important numbers before you use them.
| Check | How to see it | If it's wrong |
|---|---|---|
| Numbers | Compare a total or a few key cells with the original | Correct the cell by hand, or re-scan at a higher resolution and run OCR again |
| Column count | Count the columns in the export against the picture | Rebuild the header row and merge or split cells in the target app |
| Look-alike characters | Scan for 1 and l, 0 and O, 5 and S, or missing decimal points | Fix in the export; a clearer scan reduces these errors |
| Row completeness | Count the rows and check the first and last ones | Confirm the page was included in OCR, then run OCR on that page again |
| Negative numbers and symbols | Look for minus signs, currency symbols, and percent signs | Add them back manually, since they are easy to drop |
If several checks fail, the scan itself is usually the cause. Re-scan the page straighter and with more light, then repeat the OCR and export.
❓ FAQ
1. Can I extract a table from a photo directly?
Convert the photo to PDF first, then run OCR on that PDF in the desktop client. OCR is applied to PDF pages, so the conversion step is what makes the photo usable.
2. Why does my Excel export put all the text in one column?
The recognized text may not have a clear column structure, especially when the table has merged headers or little spacing between columns. Split the text into columns in Excel, or export to Word and rebuild the table there.
3. Do I need to run OCR on the whole document?
No. Running OCR only on the pages that contain the table is enough, and it leaves fewer pages to review afterward.
4. My scan is tilted. Should I fix it before OCR?
Yes. Rotate the page in the PDF editor before running OCR. Straight rows recognize more reliably than tilted ones.
5. Will the table borders and shading carry over?
Check the export to see what came through. Focus on whether the cells, headers, and numbers are correct, and rebuild any visual formatting you need in the target app.
📝 Summary
A table that exists only as a scan or picture needs OCR before you can copy or export it. Convert images to PDF, run OCR on the table pages, and export to Excel or Word. Then check the totals and columns against the original, because OCR can misread characters. That process is how to extract a table from image AI tools and scanned files without losing the numbers.
