Saving tables as CSV from Read by Zaweba

Plenty of the paperwork people want as text is really a table: a price list, a bank statement, a timetable, a stock sheet, results printed in rows and columns. Typing those out by hand is slow and error prone. Read by Zaweba can save them as CSV, the plain format that every spreadsheet program opens, so the rows and columns arrive ready to sort and add up.

Read calls this best effort, and that is the honest way to think of it. It works well on clean, regular tables and needs checking on anything messier.

How Read finds the cells

Read looks for rows of words that line up across wide gaps. When three or more rows align like that, it treats them as a table. In the CSV each row of the table becomes a row, and the words are split into cells at the wide gaps between them.

When no table is found on the page, the CSV still has something useful in it: one row per line of text. That way you're never left with an empty file.

Tables that work well

A typed or printed table with clear space between the columns gives the best result. Statements, price lists and simple schedules are usually like this. Ruled lines don't matter much either way; it's the gaps between the columns that Read uses.

Harder cases are tables where columns sit close together, where a cell's text wraps onto two lines, or where cells are merged across columns. Those can split in the wrong place or put a wrapped line on a row of its own. That isn't a fault you can avoid entirely with any text recognition; it is the nature of reading a table from a picture.

Before you export

Start with the best picture you can. Read's clean-up finds the page in a photo, flattens it, removes shadows, straightens it and lifts faint print, but a table photographed at an angle with the far columns blurred will still lose detail. Shoot square on and close enough that the smallest figures are sharp.

Then check the text in the side-by-side review. Numbers deserve particular care: a 5 read as an S, or a missing decimal point, matters far more in a table than in a letter. With the Tesseract engine, words under 70 per cent confidence are shaded amber, and F8 jumps to the next one. Your corrections go into the CSV when you save.

Tesseract is the engine Read chooses for tables in Auto mode, because it does layout analysis. If Windows OCR read the page, use Read > Read again with Tesseract.

After you open it in a spreadsheet

Open the CSV in your spreadsheet program and look down each column. Check that every row has the same number of cells, that headings sit over the right columns and that numbers are numbers rather than text. Totals are a good test: if the column adds up to the figure printed at the bottom of the original, the reading was right.

When a .docx is better

If you want to keep the table as part of a document, with the text around it, save as .docx instead. Read writes detected tables into .docx as real tables, alongside paragraphs and page breaks. You can save both formats from the same reading.

For a stack of statements or price lists, tick CSV on the Output tab and use the Batch tab to read the whole folder in the background, into one combined result or one set of files per picture. Read by Zaweba does all of this on your PC, which matters when the table is a bank statement.