How to Extract Tables from Images and Scanned PDFs to Excel

Use Image Table to Excel for a screenshot, scan or photo containing one table. Use PDF Table to Excel when the table is inside a PDF and you may need selected pages. A PDF with selectable text can still have a visual order that differs from its internal text order, so its workbook also needs review.

This walkthrough uses the same deliberately simple table in two forms: a PNG and an image-only PDF. It has one header row, five data rows and four columns. The known Q1 and Q2 totals are 751 and 787, which gives us useful checkpoints after extraction.

Enlarge image

The same self-authored table is supplied as PNG and as an image-only PDF, so the two workflows can be compared fairly.

Prefer the original export or scan. Keep the page upright, retain enough resolution for the smallest characters, and crop unrelated text or large empty margins. For photos, flatten the page and avoid glare, shadows and perspective distortion.

Choose the language used by most cells. Borderless tables, merged cells, multi-line values, faint rules, handwriting, currency symbols and mixed languages are harder to infer because rows and columns are reconstructed from detected positions.

• Use a sharp source instead of a screenshot of a screenshot.
• Make every header and grid boundary visible.
• Test one representative difficult page before processing a long report.
• Do not upload confidential material merely to try the workflow.

Open Image Table to Excel and use the settings shown below. The controls stack on a narrow screen, but their meaning and order remain the same.

1. 1

   Select the localized sample PNG.

   Choose mvtools-table-sample.png. Selecting it does not start processing.

2. 2

   Match the language and preprocessing.

   Choose English. Leave image preprocessing enabled for this walkthrough; compare a run without it if preprocessing makes a clean source worse.

3. 3

   Confirm whether row one is a header.

   Keep the header option enabled because Work item, Region, Q1 and Q2 are column names. Disable it when the first row is actual data or a report title.

4. 4

   Start one task and wait.

   Run extraction once and wait for completion instead of submitting the same file repeatedly.

Enlarge image

Figure 1 · Select the image, match the OCR language, then decide whether preprocessing and the header option fit the source.

This real run detected 6 rows, 4 columns and 24 cells from a 48 KB PNG, with 100.0% average confidence. That high result belongs to this clean synthetic sample; a photographed receipt or dense statement can behave very differently.

Check the preview for shifted columns, missing rows and header mistakes. Then download both formats: XLSX is convenient for spreadsheet review, while JSON exposes the extracted structure and recognition details.

1. 1

   Confirm the dimensions.

   The expected grid is 6 × 4. An unexpected row or column count is an immediate reason to inspect the source and settings.

2. 2

   Use confidence to prioritize review.

   Check small text, punctuation, identifiers and weak contrast first, even when the average is high.

3. 3

   Compare the visible preview.

   The preview should show the four headings and five work-item rows in the same order as the source.

   Enlarge image

   Figure 3 · The 20-row preview is useful for a first structural check before opening the complete XLSX or JSON.

Enlarge image

Figure 2 · The actual run completed with a 6 × 4 grid and 24 detected cells.

The PDF sample contains the same table as a single image-only page. Select the PDF, choose English, keep preprocessing and the header option enabled, and enter page 1. On a longer file, syntax such as 1,3-5 processes only those pages; leaving the field empty processes all pages within the limit.

PDF Table to Excel creates one worksheet per processed page. This run processed one page and created 1 sheet at 100.0% average confidence. A table continuing across pages therefore needs manual consolidation.

1. 1

   Test representative pages first.

   Include headers, numbers and the most difficult layout before committing to the whole document.

2. 2

   Preserve the intended page order.

   Enter pages in the order you want to process them; duplicate selections are ignored.

3. 3

   Verify page and worksheet counts.

   For this example, one processed page should produce one worksheet.

   Enlarge image

   Figure 5 · The PDF run created one worksheet for the one processed page.

Enlarge image

Figure 4 · Page 1 is selected explicitly for this one-page sample; page and range syntax is useful on longer PDFs.

Open the XLSX and compare it cell by cell with the source. Verify that Q1 totals 751 and Q2 totals 787. Also check dates, decimal and thousands separators, minus signs, leading zeroes, IDs, blank cells, column order and row alignment.

OCR returns recognized values, not the original spreadsheet logic. Rebuild formulas, data types, merged headings and formatting where needed. Keep the source beside the workbook for any important decision or publication.

Image Table to Excel accepts one JPG, PNG, TIFF, BMP or WebP file up to 25 MB and has a default five-minute timeout. PDF Table to Excel accepts one PDF up to 100 MB and 50 pages, rejects encrypted or password-protected files, and has a default ten-minute timeout. Both tools support English, Simplified Chinese, Japanese, German, French, Spanish and Portuguese.

The live tools upload source and result files to the server; they are normally cleaned up after 30 minutes. Download XLSX and JSON promptly and keep your own original. The downloadable practice files on this guide are static, self-authored public examples, not retained user tasks.