DocTable

Convert image to CSV online — real table extraction

Updated October 2026

DocTable turns a JPG, PNG or screenshot of a table into a clean CSV file with the actual rows and columns, not a pixel dump. Your first 3 pages are free, download included, and the first page needs no account; after that $3 buys 20 pages, one-time with no subscription. It reads photos and screenshots with no text layer, rebuilds true rows and columns, and opens correctly with Arabic, Persian and mixed scripts.

Start — upload an image
Item,Description,Qty,Amount
1,Consulting May,10,1200.00
2,Design review,4,480.00
3,Hosting annual,1,240.00
Sample output

How it works

  1. Upload a JPG, PNG or screenshot of a table.
  2. DocTable's AI OCR detects the rows and columns.
  3. 3 free pages, download included — see the extracted table first.
  4. Download clean CSV (or Excel) — included in your 3 free pages.

When CSV is the right output, and when it is not

CSV is the format to pick when the data is going somewhere else: an import into a database, an accounting package, a Python or R script, or a Google Sheet. It is plain text, every tool reads it, and nothing about it can break on the way in. It also carries no formatting, no formulas, no colours and no multiple sheets — so if a human is going to read the result directly, Excel is usually the better download. The same extraction produces both; you choose the format at download time.

How label/value fields and several tables are laid out

CSV is one flat grid, with no sheets to separate things into. So when a document holds both kinds of content, the export stacks them: any label/value fields come first as a two-column key,value block, then a blank line, then each extracted table with its own header row. Several tables in one document all land in the same file, one after another. That is readable and imports fine, but if you need the tables kept properly apart, the Excel export is the better choice — same extraction, different download.

Non-Latin text opens correctly in Excel

The file is UTF-8 and begins with a byte-order mark, which is what stops Excel on Windows from guessing the encoding wrongly. In practice that means an Arabic, Persian, Chinese, Japanese, Korean or Hindi table opens correctly by double-clicking the downloaded file — no Data → From Text/CSV import step and no mojibake. Google Sheets and Python's csv module read the same file without any special handling.

Why DocTable

Frequently asked questions

Can you convert a JPG to CSV?

Yes — upload a JPG or PNG and DocTable's AI OCR extracts the table into CSV rows. Your first 3 pages are free, download included, and the first page needs no account.

Is the image-to-CSV converter free?

Your first 3 pages are free — scan and download, no card needed. Your first page needs no account; sign in only to download it. After that, $3 buys 20 pages (one-time, no subscription).

Does it read the actual table, or just pixels?

The actual table — it detects rows and columns and exports structured CSV, unlike pixel-to-CSV tools.

Which image formats work?

JPG, JPEG, PNG, WEBP and screenshots; PDFs too.

How accurate is the extraction?

Accuracy depends far more on the image than on the tool. Sharp screenshots and well-lit photos read most reliably; blur, glare, low resolution and handwriting are where mistakes appear. On a 12-document test set (invoices, bank statements and receipts, including 2 degraded photos and 2 Arabic documents) DocTable read 97.5% of 521 values exactly (October 2026); real photos with blur or glare score lower, so review the extracted rows before you rely on them.

How do I make an image into a CSV file?

Upload the image, let the OCR read the table, then choose CSV at download. The file is UTF-8 with a byte-order mark, so it opens correctly in Excel by double-clicking and imports cleanly into Google Sheets, pandas or a database. Choose Excel instead if the source holds several separate tables that need to stay apart, because CSV is a single flat grid.

Related pages

Temporary processing — images are not archived. Always verify OCR output before financial use. Default OCR model is the same for every language. See Privacy in the app.