Resolution is the one scanning setting that cannot be fixed afterwards. Compression can be redone, colour can be adjusted, crooked pages can be straightened — but detail that was never captured is gone, and the original is often gone too. This is a guide to picking a number deliberately rather than accepting the scanner's default, and to understanding why "higher is better" stops being true sooner than you would think.
What dpi actually measures
Dots per inch is the sampling rate: how many pixels the scanner records per inch of original. At 300 dpi, an A4 page (8.27 × 11.69 inches) becomes roughly 2480 × 3508 pixels.
Two consequences follow immediately.
File size grows with the square. Doubling from 300 to 600 dpi quadruples the pixel count, and roughly quadruples the uncompressed size. From 300 to 1200 dpi is a factor of sixteen. This is why a "let's use the highest setting to be safe" instinct produces 400 MB scans of a twenty-page report.
Legibility does not grow with the square. Once you are sampling a letterform densely enough to reconstruct it, more samples add noise, paper texture and file size rather than information.
The numbers, by purpose
200 dpi — the floor. Adequate for large, clean printed text where the scan is a temporary convenience copy. Below this, 10-point text starts breaking down. Fax quality is around 200 × 100 dpi, which is why faxes look the way they do.
300 dpi — the working standard for document scanning, and the right default for almost everything. Ordinary printed text is fully legible, OCR accuracy is high, and file sizes are manageable. If you set one number and never think about it again, set this one.
400 dpi — a useful step up for OCR of small type (8 point and below), poor photocopies, or documents with dense tables. The size cost over 300 is modest and the OCR gain on marginal originals is real.
600 dpi — for small print, footnotes, fine detail, handwriting you need to read carefully, engineering drawings, and anything that will be enlarged. Also the sensible choice for a preservation master where the original will not survive or will not be available again.
1200 dpi and above — photographic and specialist work: halftone reproduction, film, stamps and coins, forensic examination of a signature or a paper fibre. Not for documents. At this resolution you are photographing the paper, not the text.
For photographs specifically the calculus differs: scan at a resolution matching how large you intend to reproduce the image, usually 600–1200 dpi for a small print you plan to enlarge.
Bit depth interacts with resolution
Resolution is only half the sampling decision. The other half is how much information each sample carries.
- Bilevel (1-bit): each pixel is black or white. Tiny files, and it destroys anything faint — pencil, light stamps, highlighter, halftones. At low resolution bilevel is brutal, because thresholding a partially-covered pixel throws away the anti-aliasing that made the letterform readable. If you must use bilevel, use it at 600 dpi rather than 300.
- Greyscale (8-bit): the right default for text documents. Handles faint marks, survives compression well, OCRs better than bilevel on imperfect originals.
- Colour (24-bit): three times greyscale in size. Use it when colour is information — signatures in blue ink proving they are not photocopies, colour-coded forms, charts, photographs, letterheads that matter.
A common and effective combination is 300 dpi greyscale for routine documents and 600 dpi colour for anything being preserved.
Effective dpi, and why it is the number that matters
The dpi you scan at is not always the dpi that ends up in the document. Two things change it:
Downsampling during optimisation. Every "compress PDF" tool has a target resolution, frequently 150 dpi. Run a 600 dpi scan through it and you have a 150 dpi scan in a smaller file. See lossy vs lossless PDF compression.
Placement scale. If an image is placed on a page larger or smaller than its natural size, the effective resolution changes accordingly.
To see what a PDF actually contains:
pdfimages -list document.pdf
The x-ppi and y-ppi columns show the effective resolution of each image as placed. This is the honest measure, and it is frequently a shock — many "300 dpi archives" turn out to contain 150 dpi images after somebody optimised the folder. Tooling background: poppler-utils introduction.
Resolution and OCR
OCR engines are trained on particular character heights in pixels. Tesseract, the most widely used open engine, performs best when lowercase letters are roughly 20–30 pixels tall; below about 10 pixels accuracy collapses.
Working backwards: 10-point text has a lowercase x-height of about 1.5 mm. At 300 dpi that is roughly 18 pixels — comfortably in range. At 150 dpi it is 9 pixels, right at the failure boundary. This is the concrete reason 300 dpi is the standard and 150 dpi produces the frustrating half-accurate OCR people complain about.
For 6-point footnote text, apply the same arithmetic and you land at 600 dpi. More on the engine side in OCR language support and accuracy and PDF OCR explained.
Two related points: OCR generally does not benefit from scanning above 600 dpi — accuracy plateaus and then declines slightly as paper texture becomes visible as noise. And greyscale beats bilevel for OCR on anything less than pristine, because the engine can use the intermediate tones.
Size expectations
Rough figures for a single A4 page, after normal compression:
- 300 dpi bilevel, CCITT G4: 30–80 KB
- 300 dpi greyscale, JPEG: 150–400 KB
- 300 dpi colour, JPEG: 300 KB–1 MB
- 600 dpi greyscale: 500 KB–1.5 MB
- 600 dpi colour: 1.5–4 MB
Multiply by page count and you can predict the problem before it happens. A 200-page colour scan at 600 dpi is somewhere around 500 MB, which is a storage and transfer decision, not an accident. If you land there and need it smaller, how to compress a scanned PDF and why is my PDF so large cover the options — but the better answer is usually to have scanned differently.
The preservation master pattern
For anything genuinely worth keeping, the archival world's standard approach avoids the trade-off entirely: scan once at high quality, derive everything else.
- Master: 600 dpi colour, lossless or lightly compressed, kept in cold storage and never edited.
- Access copy: 300 dpi greyscale, OCR'd, compressed, this is what people actually open.
- Thumbnail: whatever the catalogue needs.
The master costs storage, which is cheap; rescanning costs handling the original again, which is expensive and sometimes impossible. The framework this comes from is described in the OAIS model for document preservation, and the practical side in how to archive PDFs long term.
Quick decision table
- Everyday office document, will be read and filed → 300 dpi greyscale.
- Receipt, thermal paper, faint print → 400 dpi greyscale.
- Contract with signatures you want to show as genuine → 300 dpi colour.
- Small print, footnotes, dense tables, serious OCR → 600 dpi greyscale.
- Historical document, one-time access to the original → 600 dpi colour, kept as a master.
- Photograph for reuse → 600–1200 dpi colour.
- Drawing that will be enlarged or measured → 600 dpi minimum.
- Throwaway copy to email in the next five minutes → 200–300 dpi, whatever is fastest.
Summary
Default to 300 dpi greyscale; step up to 600 for small type, handwriting, or preservation; use colour only when colour carries meaning. Verify with pdfimages -list that what reached the file is what you scanned, because optimisation tools quietly downsample. And when the original is fragile or about to be destroyed, scan high once and derive the working copies — that is the only decision here you cannot revisit.
