What this covers
- Divide the size by the page count - it tells you almost everything
- Under about 150 KB per page means text; over 500 KB means images
- Compression only acts on images, so a text document barely shrinks
- Rescanning correctly beats compressing a bad scan
A PDF is a container, and how large one is depends almost entirely on what was put in it. The question is never really 'why is this file large' but 'what is in it that weighs anything', and there is usually one answer.
Divide by the page count
This single calculation identifies the situation in seconds. A generated document of text and vector graphics runs 30 to 150 kilobytes per page. A scanned page runs from about 200 kilobytes to several megabytes.
So a 40-page document at 3 megabytes is generated and entirely normal. A 40-page document at 90 megabytes is a colour scan. A 6-page document at 40 megabytes has photographs in it. The arithmetic is cruder than any analysis tool and it is right nearly every time.
What weighs what
Measured across a set of real documents.
| Content | Per page | Notes |
|---|---|---|
| Plain text | 20-60 KB | Mostly the embedded fonts |
| Text with vector diagrams | 50-150 KB | Vectors are remarkably cheap |
| A greyscale scan at 200 DPI | 100-300 KB | The sensible target for a scan |
| A colour scan at 300 DPI | 1-5 MB | The usual cause of an enormous file |
| An unprocessed phone photograph | 2-5 MB | Before any clean-up |
| A page with one high-resolution photograph | 500 KB-3 MB | Depends entirely on the photograph |
Why compressing a text document does almost nothing
Compression recompresses embedded images at a lower resolution and quality, and rebuilds the file structure to drop anything unreferenced. On a document whose weight is images, it is very effective. On a document of text and vector diagrams, there is nothing image-shaped to act on.
A five per cent saving on a text document is not a failure. It is the correct answer, and it tells you the file was already close to as small as that content can be. Any service reporting a large reduction on such a file is rasterising the text - which makes it blurry, unsearchable and usually larger.
The honest expectation
Colour scan: seventy per cent or more. Greyscale scan: thirty to fifty per cent. Text document with a few images: ten to twenty per cent. Pure text and vectors: under ten, and often under five.
Finding out what the weight is
Two minutes.
- Divide the size by the page count
This usually settles it on its own.
- Try to select text
If you cannot, the whole document is images and that is the entire answer.
- Run image extraction
It reports how many embedded images there are and how large the result is. A handful of very large images is a different problem from a scan.
- Check for pages you do not need
Blanks and duplicated inserts are free to remove, and on a scanned set can be a third of the file.
Investigating and fixing
The single oversized image
A document that is mostly text but unexpectedly large usually has one culprit: a photograph or a screenshot inserted at full camera resolution, scaled down to occupy a quarter of a page.
The page displays it small; the file stores it at its original size. Image extraction finds it immediately - it will be conspicuously larger than everything else. Compression handles it well, because resampling a 4,000-pixel image down to the 800 pixels the page actually displays is exactly what compression does.
When the answer is to rescan
If the document is a scan and compression has not got it small enough, the capture was the problem rather than the compression. A colour photograph of a page taken in dim light contains a lot of noise, and noise is expensive to store and compresses badly.
Reshooting in better light and converting to greyscale produces a smaller and more readable document than any amount of processing the original. This is the case where doing the work again is genuinely faster than fixing it.
Two 40-page documents
Same page count, entirely different files.
3 MB
- Generated from a word processor
- Text with embedded fonts, a few vector charts
- Fully searchable
- About 75 KB per page
- Compression will save almost nothing
90 MB
- Scanned in colour at 300 DPI
- Forty photographs of pages
- No text layer at all
- About 2.3 MB per page
- Compression will save seventy per cent or more
Diagnosing an oversized PDF
- Divide the file size by the page count
- Try to select text to establish whether it is a scan
- Extract the images to find one oversized culprit
- Delete blank and duplicated pages first
- Compress, and read the reported saving rather than assuming
- For a scan that is still too large, rescan rather than recompress
Questions and answers
Why did compression barely shrink my PDF?
Because its weight was text and vector drawing rather than images. There was nothing for compression to act on, and that small saving is the correct result rather than a failure.
How large should a PDF be?
Roughly 30 to 150 KB per page for a generated document, and 100 KB to several MB per page for a scan. Divide the size by the page count and compare.
Can I make a scanned PDF as small as a text one?
Close, at the black-and-white setting, which stores one bit per pixel and compresses extraordinarily well. It erases faint pencil, so check the result before relying on it.
One page is much larger than the others. Why?
It almost certainly contains a full-resolution photograph or screenshot displayed at a small size. Image extraction will identify it, and compression resamples it effectively.
