OCR for tables: turning grids into Markdown you can trust

A misread word in a paragraph is annoying. A misread digit in a financial table can be expensive. Tables are high-value OCR targets and high-risk ones: structure and characters both have to survive.

Textify Image reconstructs tables into Markdown pipe tables so you can paste them into docs or clean them up quickly. This guide focuses on getting grids right.

Why tables are hard

OCR must simultaneously:

1. Find cell boundaries
2. Read text inside each cell
3. Keep rows and columns aligned
4. Handle merged cells, ruling lines, and missing borders

Printed spreadsheets with clear black gridlines are easier than minimal “whitespace only” layouts. Handwritten tallies are harder still.

Capture tips for tabular pages

• Fill the frame with the table; avoid capturing half of the next table on the page.
• Keep horizontal rules as horizontal as possible — rotate the photo before upload if needed.
• Ensure the rightmost column is not cut off (a common phone mistake).
• For very wide tables, photograph left and right halves with a shared key column (like a name or ID) so you can join them later.

Scanned PDFs of statements often beat phone photos. Prefer the scan when you have it — see extract text from PDFs.

Reading Markdown tables

A simple result looks like:

| Item | Qty | Price |
| --- | --- | --- |
| Paper | 12 | 4.50 |
| Toner | 2 | 39.00 |

Rules of thumb:

• The separator row (| ---) marks the header.
• Empty cells still need pipes so columns stay aligned.
• If a cell contains a pipe character, it may need escaping or rephrasing.

Verification workflow (do this for money and grades)

1. Row count — does the Markdown have the same number of data rows as the source?
2. Spot-check — verify first row, last row, and one middle row against the original.
3. Totals — if the source has a sum row, recompute in a spreadsheet.
4. Units — watch for O vs 0, l vs 1, and lost decimal points.

Only after checks should you treat the table as data.

Merged cells and multi-line headers

Real reports love merged title cells (“FY2025 actuals” spanning four columns). OCR may:

• Repeat the title in each column
• Place the title in a row above the table
• Drop the merge and only keep child headers

None of these is automatically “wrong,” but you should reshape the Markdown to match how you need the data.

From Markdown to a spreadsheet

Options:

• Paste into an editor, convert pipes to CSV with a small script or online helper you trust
• Copy cell-by-cell for tiny tables
• Rebuild headers manually and paste columns if the table is wide and messy

For critical data, a minute of structured cleanup beats silent numeric errors.

When to split the job

If a page has a narrative section and a dense appendix table, crop the table to its own image and convert separately. You will get cleaner grids and less mixed-in prose.

Related guides

• Photograph documents for better OCR
• Why Markdown structure matters
• Common OCR mistakes

Treat table OCR as “draft data plus a verification step.” Layout-aware Markdown gets you most of the way there; your eyes protect the last mile.