DEV Community

NullPointerZen
NullPointerZen

Posted on

Getting a table out of a screenshot and into a spreadsheet

I keep getting tables that are actually images — a screenshot of a schedule, a photo of a grade sheet, a report cropped out of a PDF — and I need them in a spreadsheet to keep working. For a long time that meant retyping cell by cell, going cross-eyed by row thirty. Turns out you don't have to: OCR can hand the table back with its structure intact.

I'm not an engineer, just how to use it. Drop the table image into ImgIng's (imging.ai) text recognition, pick a tier, and it reads it locally; the right side gives you a version laid out in the original rows and columns. The key is the export: the "copy with layout" and TXT outputs separate table columns with Tab.

Tab-separated is the whole point. Paste that TXT straight into Excel and each Tab jumps to the next cell, so the columns from the image land on the columns in the sheet — no manual "text to columns". First time I tried it I braced for a config screen; it was just copy, paste, and the table arranged itself.

And it's not Excel-only: WPS, Numbers, Google Sheets, Feishu Sheets all split on Tab the same way, so whatever spreadsheet you use, the path works. What I do most is move a chunk of data cropped from a PDF report into a sheet to keep calculating — used to be half an hour of typing, now it's recognise-and-paste in under two minutes.

A picking tip: tables with clear gridlines and short cell contents read best, Tab-splitting almost never errs; borderless tables aligned only by whitespace, or a cell stuffed with a whole paragraph, make column inference struggle — check the columns against the original after. The more regular the image, the smoother this goes.

Two spots you must check before using it, though. Merged cells — those big boxes spanning two rows or columns — trip up column inference, since it works from box geometry; it can split them wrong or shift a row, and the tool itself says to verify row-spanning merges. And low-confidence characters, digits especially: a 0 read as 8, a 1 as 7, and one wrong number poisons the whole sheet. Those get flagged in the output, and a one-minute scan beats reworking it later.

So the whole flow is three steps: drop the table image in, paste the TXT into Excel and let Tab split the columns, then check merged cells and numbers. Far faster than retyping, and the image never left my machine — I checked, it marks the work local.

Top comments (0)