Documents and AI
Scanning a document properly
Five minutes of method that avoid half the reading problems.
The quality of the original decides almost everything that follows: whether it reads well, whether you can search inside it, whether it has to be reviewed by hand. And fixing it costs five minutes at scanning time, against hours of review afterwards.
The numbers that work
| Setting | Value | Why |
|---|---|---|
| Resolution | 300 dpi | Enough to read; 600 weighs four times as much and gains nothing |
| Colour | Greyscale | Unless colour means something, such as a stamp |
| Format | Better than loose images: it keeps the page order |
If you use a phone
A current phone in decent light matches a scanner. What ruins photos is not the camera, it is hurry.
Watch out
If the paper is creased or the print is very faint, no scan will fix it. What is needed there is another original.
Worth knowing
A document that already exists digitally should not be scanned: ask for the PDF. Scanning a printout of a PDF loses the text it already contained, which then has to be recognised by OCR.
›What about fifty pages?
In one PDF, not fifty images.
›Is scanning in colour worth it?
Only if colour adds something: a stamp, a blue signature, a chart.
›Can I crop the photo?
Yes, to remove the table. Do not retouch anything else if the document may serve as evidence.
A real case
The situation
A supplier's certificates read badly again and again.
What you do
- The original is checked: hand-held photos with shadow
- This guidance is sent to them
What you get
From the next send onward reading comes back high-confidence and manual review stops.
The situation
The scan comes out dark and cannot be read.
What you do
- Redoes it with more light and no shadows
What you get
Reading works first time.
The situation
It is scanned at very low resolution to keep it small.
What you do
- Raises the resolution without exceeding the limit
What you get
The document goes in and also reads.
The situation
The document comes out cropped at the edges.
What you do
- Frames the whole document
What you get
No text is missing from the reading.
The situation
Black text is scanned in colour.
What you do
- Scans in whichever mode gives the best contrast
What you get
The file is smaller and reads better.
The situation
Each person scans differently.
What you do
- Agrees common settings
What you get
Quality stops depending on who scans.
This article answers
- how to scan a document so it reads well
- what dpi to scan at
- photograph a document with a phone
- the scan is too large