DPI
Dots per inch · PPI
Dots per inch: how many pixels represent one inch of a page when it is scanned or rendered. It sets how many pixels each letter gets.
In DocSlurpGlossary entry
On this page
DPI (dots per inch) is the number of pixels used to represent one inch of a page when it is scanned or rendered. It is the single biggest knob on what OCR and vision models can see.
Type is measured in points, 72 to the inch, so the arithmetic is simple: pixels = points × DPI / 72. An 11-point line of type gets:
| DPI | Pixels for 11 pt | US Letter page |
|---|---|---|
| 72 | 11 | 612 × 792 |
| 150 | about 23 | 1,275 × 1,650 |
| 300 | about 46 | 2,550 × 3,300 |
Lowercase letters are roughly half that height, so at 72 DPI a lowercase “e” has about 5 pixels to work with. That is not enough to tell “e” from “c” reliably. Tesseract’s documentation recommends at least 300 DPI for scanned input.
Higher is not free. Doubling DPI quadruples the pixel count, and for a vision-language model more pixels usually means more image tokens and a higher bill. Many pipelines render born-digital pages at a moderate resolution and reserve high resolution for small print or degraded scans.
The gotcha: DPI cannot add detail that was never captured. Upscaling a 100 DPI fax to 300 DPI makes a bigger, blurrier image, not a sharper one. The resolution that matters is the one at which the page was scanned. See rasterization.
In DocSlurp
In DocSlurp
Page images render at 150 DPI by default, clamped to 72–300, and capped so no side exceeds about 4,000 pixels and no page exceeds about 16 megapixels (core/worker/src/pipeline/pdf-asset-cache.ts, core/worker/src/lib/raster-bounds.ts). The hidden-OCR-layer check renders a band at 100 DPI (core/worker/src/routing/probe/ocr-layer.ts).