This route produces a plain-text OCR draft beside the source. It does not claim automatic cell-perfect Markdown; the guide shows how to reconstruct a valid table without hiding uncertain relationships.
This tool intentionally returns a text draft. The page earns its route by showing how to reconstruct Markdown without pretending that plain OCR preserved merged cells or column relationships.
Recognition is only the first stage. Use Table OCR Quality Control: Rows, Columns and Merged Cells for the relevant control checks and Numeric OCR Validation for Dates, Totals and Identifiers for values that must be exact. If the output structure changes, compare Chart Image to Data with Invoice and Receipt OCR.
Validate the table as a grid
| Review point | Likely failure | Validation action |
|---|---|---|
| Headers | Values detach from multi-level headings | Map every spanning and nested header before transcription |
| Empty cells | Blank, zero, and not applicable collapse together | Define explicit states before building the Markdown |
| Numbers | Signs, separators, or units move columns | Check each numeric cell with its row and column labels |
Rebuild relationships the text stream cannot preserve
Crop to the table while retaining its title and units.
Identify the header hierarchy before rebuilding rows.
Verify every numeric cell, blank, sign, and unit against the image.
Human-reviewed Markdown scenario
For a price table with a category label spanning three rows, decide whether to repeat the verified category in each Markdown row or use HTML with a row span. Record that transformation. Never fill an empty price from a neighboring pattern just to make the table look complete.
When the table needs coordinates or spreadsheet tools
Use a spreadsheet or layout-aware extraction tool when coordinates and cell boundaries are required at scale. Use HTML rather than Markdown when meaningful row spans or column spans must be preserved.