Structural or rasterised: choosing a PDF compression mode
The tool offers two PDF modes and one of them warns me about losing things. Which do I want?
The two modes in one sentence each
A structural save rewrites the same document more efficiently: same pages, same text, same fonts, fewer wasted bytes. A rasterised compression renders each page to a bitmap and builds a new PDF whose pages are those bitmaps: same appearance, at a chosen resolution, and nothing else.
The first is a housekeeping operation. The second is a conversion, and it is a one-way door.
What survives each mode
| Feature | Structural save | Rasterised |
|---|---|---|
| Selectable and searchable text | Kept | Lost |
| Embedded fonts | Kept or subset | Not applicable — pages are images |
| Hyperlinks | Kept | Lost |
| Bookmarks and outline | Kept | Lost |
| Form fields | Kept | Flattened into the picture |
| Annotations and comments | Kept | Flattened or dropped |
| Tagged structure for screen readers | Kept | Lost |
| Digital signatures | Invalidated by any rewrite | Lost |
| Vector sharpness when zoomed | Kept | Limited by the render resolution |
| Typical size reduction | Small to moderate, and highly variable | Large on scans, moderate elsewhere |
When structural is the only defensible choice
- Anything a person will search inside: contracts, manuals, reports, research papers.
- Anything that has to remain accessible. The tagged structure that lets a screen reader announce headings, reading order and table cells does not survive rasterisation, and a picture of a page is unreadable to assistive technology.
- Anything with working form fields that someone still has to fill in.
- Anything containing vector artwork that people zoom into: drawings, maps, plans, charts.
- Anything you may need to edit later.
In these cases, if the structural save does not get you where you need to be, the right move is usually to reduce what you are sending — split out the pages that matter — rather than to reach for the destructive mode.
When rasterising is genuinely the right answer
Rasterising makes sense when the document is already a set of pictures, which is exactly what a scan is. A scanned contract has no text layer to lose, no links, no tags and no form fields. Re-rendering its pages at a lower resolution attacks the only thing that is large, and costs you nothing that was there in the first place.
It is also reasonable for a document whose only remaining job is to be looked at once: a receipt for an expense claim, a proof of address for a portal with a hard size cap, a poster for a notice board. Just be certain that is genuinely the document's only remaining job.
Resolution is the whole quality argument
When you rasterise, the number that decides whether the result is usable is the render resolution in dots per inch. Text on paper is legible on screen at surprisingly modest resolutions, and unreadable just below them. Small print, footnotes, superscripts and fine tables are the first casualties, and they are also the parts people most often need to read.
| Render DPI | Approximate page pixels | Practical use |
|---|---|---|
| 72 DPI | 595 × 842 | Screen thumbnail; body text becomes hard work |
| 96 DPI | 794 × 1123 | On-screen reading of ordinary body text |
| 150 DPI | 1240 × 1754 | Comfortable reading, acceptable for most printing |
| 300 DPI | 2480 × 3508 | Print-quality; usually the resolution a scan arrived at |
Rendering above the resolution the source images already have adds bytes and no detail. If a scan came in at 200 DPI, rasterising it at 300 produces a bigger file containing the same blur.
The one-way door
There is no un-rasterise. Once the text layer is gone, the only way back is optical character recognition, which is a guess at what the pixels say and introduces its own errors. Keep the original file. Treat the rasterised copy as a disposable artefact made for one delivery, in the same way you would treat a JPEG exported from a raw photograph.
Signatures deserve a final warning of their own. Any modification to a signed PDF, structural or rasterised, breaks the cryptographic relationship between the signature and the bytes it covers. Viewers will report the signature as invalid, which for many documents is worse than the file being large.
Deciding in thirty seconds
- Try to select a sentence with the cursor. If you cannot, it is a scan — rasterising is safe and effective.
- If you can select text, ask whether anyone will ever search, copy, click or read this with assistive technology. If yes, stay structural.
- If the structural result is not small enough, split the document or extract the pages you actually need.
- Only then consider rasterising, and only on a copy.
Tools for this
Sources
- ISO 32000-2 — Document management, Portable Document Format
- W3C — PDF techniques for WCAG, including tagged structure
- ISO 19005 — PDF/A, the archival profile that requires text and structure