What kinds of tables work best?

Simple grids: one value per cell, consistent alignment, visible spacing between columns. Clean screenshots of digital tables work especially well. Merged cells, multi-line cells, and nested headers will need manual fixing after export.

Two columns came out merged into one — why?

Column boundaries are detected from horizontal gaps, so if two columns sit closer together than the normal spacing between words, the tool can't tell them apart. A higher-resolution image helps, as does cropping tightly to the table.

Can it read a handwritten table?

Realistically, no. Tesseract is weak on handwriting, and tabular handwriting compounds the problem — one misplaced word corrupts the row structure. For important handwritten data, manual entry beats fixing the output.

Why CSV instead of an Excel file?

CSV opens in every spreadsheet application and pastes cleanly into an existing sheet, which is usually the destination anyway. Once it's in Excel or Sheets you can save it as .xlsx.

Can I extract a table from a PDF?

Not directly — this tool accepts images only. If the PDF is a scan, convert the page with PDF to Images first. If the PDF has real selectable text, skip OCR and use the PDF to Excel tool, which reads the text directly and is far more accurate.

How careful do I need to be with the numbers?

Careful. OCR confuses similar glyphs — 0 and O, 1 and l, 5 and S — and in a data table those errors are silent. Spot-check against the original, or recompute a column total as a sanity check.

Table OCR

Extract table data from images into CSV format using OCR.

Drop image files here or browse

Max 50MB per file

Language

About Table OCR

Extracting a table from an image is harder than ordinary OCR: you don't just need the words, you need to know which row and column each word belongs to. Table OCR recognizes the text in a photographed or scanned table, records the position of every word, clusters those positions into rows and columns, and writes the result out as CSV — ready for Excel or Google Sheets.

The clustering is geometric. Words whose vertical centers line up are grouped into a row; within each row, gaps clearly wider than the page's typical word spacing become column boundaries. That's why the tool works best on tables with consistent spacing, and struggles when columns nearly touch.

Like every Convertora tool, it runs fully in your browser with Tesseract.js: the language model downloads once, and your image never leaves your device — relevant when the table is a payroll sheet or lab result.

How to use it

1Upload an image of the table — JPG, PNG, or WebP, up to 50 MB. Images only: for a table inside a PDF, screenshot the page or convert it with PDF to Images first.
2Pick the language of the table's text (English, Spanish, French, German, Chinese (Simplified), Japanese, Korean, Portuguese, Italian, or Dutch).
3Click Extract Table. The engine recognizes every word with its position, sorts words into rows top to bottom, and splits each row into cells at the wide gaps.
4Review the CSV in the output box — rows are padded to the same column count, and cells containing commas are quoted properly.
5Copy the CSV to your clipboard or download it as a .csv file, then open or paste it in your spreadsheet.

Common use cases

Getting a supplier's price list or rate card into a spreadsheet you can sort and compare.
Digitizing a printed financial statement, timetable, or inventory sheet without retyping every cell.
Pulling a results table out of a scanned report or academic paper for your own analysis.
Converting a photographed whiteboard or printed schedule into a working spreadsheet.

Frequently asked questions

Tips

Photograph or scan the table straight-on — perspective distortion bends the rows, and row clustering assumes cells line up horizontally.
Crop the image to just the table; captions, page numbers, and surrounding paragraphs get pulled into the grid as extra rows.
If a complex table exports badly, split the image into simpler sections and recombine in the spreadsheet.
Check the header row first after import — if the headers landed in the right columns, the data below usually did too.

100% private — runs in your browser

Convertora processes everything on your device using JavaScript and WebAssembly. Files never leave your browser, are never uploaded to a server, and are never seen by us or anyone else. The moment you close the tab, the data is gone — there is no temporary cloud copy, no log entry, no retained backup.

Because the work happens locally, processing speed depends on your device — but there are no rate limits, no daily caps, and no file size restrictions beyond what your browser can handle in memory. No signup, no account, no payment. The tool works the same in incognito mode, on a corporate network, or after the page has loaded once, even with the network disconnected.