What DPI to scan documents: text, OCR and archives
Scan documents at 300 DPI, the figure text recognition and the US federal guidelines ask for. When 200 will do, when to use 400 or 600, and the file sizes.
Published
Scan documents at 300 DPI. It is the resolution the US federal guidelines set for modern office records, it is the minimum the Tesseract text-recognition engine recommends, and it reprints at the original size at full print quality. Drop to 200 only for papers you just need to be able to read. Go up to 400 or 600 for small print, fine detail, and anything you may want to enlarge.
Photographs are a separate question with higher numbers: see what DPI to scan photos.
The short table
| What you are scanning | DPI | Why |
|---|---|---|
| Letters, bills, forms, typed and printed pages | 300 | The federal figure for modern text; reliable for text recognition |
| Papers to read once and file | 200 | Legible and less than half the data of 300 |
| Small print, footnotes, faint carbon copies, stamps | 400 to 600 | More pixels in each small letter |
| Line drawings, plans, signatures to reuse | 600 | Hard edges show the pixel grid first |
| Old or valuable manuscripts | 400 | The top level in the federal guidelines for rare documents |
| Anything to be printed larger than the original | 300 × the enlargement | A page printed at twice the size needs 600 |
Where 300 comes from
The archives. The Federal Agencies Digital Guidelines Initiative publishes the Technical Guidelines for Digitizing Cultural Heritage Materials (third edition, 2023), which US agencies digitise to. It grades scans from one to four stars and gives a resolution for each kind of paper:
| Kind of document | 1 star | 2 star | 3 star | 4 star |
|---|---|---|---|---|
| Unbound documents, general collections | 200 ppi | 250 ppi | 300 ppi | 400 ppi |
| Manuscripts and other rare materials | not offered | 250 ppi | 300 ppi | 400 ppi |
| Modern textual records | 300 ppi, no star levels |
“Modern textual records” are what most households and offices have: in the guidelines’ words, “paper records with well-defined printed type (such as typeset, typed, laser-printed, etc.), and with moderate to high contrast between the ink of the text and the paper background”. For those there is one figure, 300, in colour or greyscale.
The guidelines describe three stars as “a very good professional image that is appropriate for most uses” and one star as fit “to provide a reference to locate the original”. So 200 is a reference copy, 300 is a good copy, and 400 is the best an archive aims for with paper. On going further they say: “Image capture resolutions above 400 ppi may be appropriate for some materials”, and for maps and posters, “very fine detailed engravings, for example, may require 800 ppi or higher”.
Text recognition. If the scan will be made searchable, the recognition software sets a floor. The documentation for Tesseract, a widely used open-source recognition engine, says: “Tesseract works best on images which have a DPI of at least 300 dpi” (Improving the quality of the output). The federal guidelines agree in their own terms: three-star imaging “generally produces image files capable of high quality conversion”, though “testing should be done” because software and paper vary.
Reprinting. Print is judged at 300 pixels to the inch, so a 300 DPI scan reprints at its own size. See what DPI for printing.
What each setting costs
A scan’s pixel count is the paper size in inches times the DPI. Uncompressed, greyscale takes one byte a pixel and colour three:
| DPI | US Letter in pixels | A4 in pixels | Megapixels | Greyscale, raw | Colour, raw |
|---|---|---|---|---|---|
| 150 | 1275 × 1650 | 1240 × 1754 | 2.1 | 2 MB | 6 MB |
| 200 | 1700 × 2200 | 1654 × 2339 | 3.7 | 4 MB | 11 MB |
| 300 | 2550 × 3300 | 2480 × 3508 | 8.4 | 8 MB | 24 MB |
| 400 | 3400 × 4400 | 3307 × 4677 | 15.0 | 14 MB | 43 MB |
| 600 | 5100 × 6600 | 4961 × 7016 | 33.7 | 32 MB | 96 MB |
The megapixel and size columns are for US Letter; A4 is within a few per cent. Saved files are far smaller than the raw figures, because PDF, PNG and JPEG all compress, but the proportions hold: 300 has two and a quarter times the pixels of 200, and 600 has four times the pixels of 300.
200 DPI vs 300 DPI
The difference shows in small letters. Type is measured in points, 72 to the inch, so the height of a line of type in pixels is its point size × DPI ÷ 72:
| Type size | At 150 DPI | At 200 DPI | At 300 DPI | At 600 DPI |
|---|---|---|---|---|
| 10 point body text | 21 px | 28 px | 42 px | 83 px |
| 8 point | 17 px | 22 px | 33 px | 67 px |
| 6 point small print | 13 px | 17 px | 25 px | 50 px |
Those are the heights of the full body of the type; the letters themselves are smaller, and a lower-case “e” is roughly half of it. At 200 DPI, 10 point text is comfortable to read on screen and prints a little soft. Six point small print at 200 has letters under ten pixels tall, which leaves little to tell an “e” from a “c”. Recognition depends on how many pixels each letter gets: Tesseract’s documentation points to tests of the best height of a capital letter in pixels. At 300 the same small print has half as many pixels again in each direction.
So: 200 for reading, 300 for keeping, searching and reprinting.
Colour, greyscale or black and white
- Greyscale suits most text. It keeps pencil, faint stamps and the texture of the paper, at a third of the raw size of colour.
- Colour when colour carries meaning: letterheads, highlighter, coloured ink, stamps, old paper whose tone you want to keep. The federal figure for modern text allows “color or grayscale”.
- Black and white (1-bit) makes the smallest files, but every pixel is forced to pure black or pure white, so faint marks vanish and edges turn jagged. If you use it, use 600 DPI: line art is the case where a book printer such as IngramSpark asks for 600 ppi, against 300 for greyscale, in its File Creation Guide.
Scan in greyscale or colour first. Recognition software does its own conversion to black and white. Tesseract’s documentation notes that the result “can be suboptimal, particularly if the page background is of uneven darkness”, and a greyscale original leaves you room to correct that.
Getting a clean scan
Resolution is not the only thing that decides whether text is recognised. From Tesseract’s documentation:
- Straighten the page. “The quality of Tesseract’s line segmentation reduces significantly if a page is too skewed”.
- Dark text on a light background.
- Trim dark scanner borders, which “can be erroneously picked up as extra characters”, but leave a small margin round the text.
And from the federal guidelines: scan the whole sheet. Documents “should be imaged to include the entire area of the page and a small amount beyond to define the page area”.
Fragile papers
The US National Archives’ advice on digitising family papers:
- “Automatic feed scanners are not suitable for fragile, weak, bent, or valuable papers; papers can jam and become torn in automatic feed scanners.” Use the flatbed glass for anything old.
- Make sure the paper fits on the glass: “The lid of the scanner can crush and crease the original if the paper doesn’t fit on the scanner.”
- For books, “use a copy stand instead of a flatbed scanner”, so the binding is not forced flat.
- Keep the originals, and keep “three copies, stored on two different media, and one copy located off-site”.
Check the result
If the scanner saved a PNG or a JPEG, the image DPI checker shows its pixel size, the DPI written in the file and the size it prints. A Letter page scanned at 300 should read 2550 × 3300 pixels, and an A4 page 2480 × 3508. If it reads half that, the scanner software was set lower than you thought.
A scanned PDF holds the same page images inside it. The checker reads PNG and JPEG files, not PDFs; to judge a PDF, zoom to 400% and see whether small letters stay crisp.
Questions
Is 600 DPI too much for documents? For ordinary text, yes: four times the data of 300 for no gain in reading. It earns its place for small print, line art and pages you will enlarge.
Is 150 DPI enough? For a reference copy of large, clear type. Small print becomes hard to read and text recognition suffers.
What DPI for scanning to email? No source sets a figure for this. Our suggestion is 200: ordinary type stays legible, and the scan has less than half the pixels of one at 300, so the attachment is smaller. Use 300 if the recipient may print it or needs to search the text.
What DPI for scanning receipts? Our suggestion is 300, because receipt type is small and often faint, and 300 is the figure Tesseract’s documentation gives if the text is to be recognised.
Does a higher DPI make text recognition better? Tesseract’s documentation says it works best at 300 DPI or more, and gives no promise beyond that. A straight, clean, evenly lit page matters as well.