PDF Compression Explained

Where the bytes go

Every PDF is a set of objects: page content streams (the drawing instructions), fonts, images, annotations, forms and a table that says where each object lives. A size analysis lists these by category and usually tells the whole story in one look.

• Images — by far the usual culprit. A scanned page at 300 dpi is a 2,500 × 3,500 pixel picture; stored as a lossless PNG-style image it can take 5–20 MB per page.
• Fonts — a full Unicode font is 2–20 MB. Well-made PDFs embed only the glyphs they use (a "subset"); many exporters embed the whole font, once per document section.
• Incremental updates — each save in some editors appends a new revision rather than rewriting the file, so an edited document carries every previous version.
• Duplicate objects — the same logo image or font embedded once per page instead of once per file.
• Extras — thumbnails, embedded files, JavaScript and metadata streams that are not needed to display the pages. Structure tags do not belong on this list: screen readers and reflow depend on them (tagged PDF, PDF/UA), so a compressor should keep them.

Try it: PDF Size Analyzer Try it: PDF Font Analyzer Try it: PDF Image Extractor & Analyzer

Lossless: what changes nothing you can see

Lossless steps rewrite the file without touching a single pixel or glyph. Object streams pack small objects together and deflate them; a fresh cross-reference table drops the leftovers of old revisions; identical objects are merged; unused resources are removed; content streams that were stored uncompressed are deflated. On a document that was saved carelessly, this alone can halve the size. On one that was exported well, it saves a few percent — and a good tool reports "no reduction" rather than pretending.

Font subsetting is lossless for the reader but not always for the file's future: a subset font cannot be used to type new text in that PDF later. If a form still needs to be filled, keep the full fonts.

Try it: Compress PDF Try it: PDF Cleanup & Sanitizer

Lossy: images are where the real savings are

Two levers apply to every image in the file. Resolution: a photo displayed 10 cm wide needs about 150 pixels per inch on screen and 300 for print, so a 600-dpi scan can be downsampled to a quarter of its pixels with no visible change. Encoding: re-encoding as JPEG at quality 75–85 keeps photographs looking identical at normal zoom while dividing their size by five or more. Line drawings, screenshots and text scans are the exception — JPEG blurs sharp edges, so they should stay lossless or be kept at high quality.

A compressor that works in your browser applies these levers to each image and shows the before and after sizes. In this site's Compress PDF tool they are two settings of the "Re-compress images" mode: Image quality (JPEG quality, 70 by default) and "Downscale images larger than" (a maximum size in pixels, 1800 by default, applied to the longest side of each image — about 150 dpi along the long side of an A4 page). As a rough guide, quality 70 with 1200 px suits email and the web, quality 80 with 1800 px suits reading on tablets, and "Keep original size" with quality 90 suits a file that will be printed. Its "Rasterise pages" mode is different: it turns every page, text included, into a JPEG at a chosen dpi — the grayscale tool's whole-page mode does the same — so text is no longer sharp, selectable or searchable. Nothing is uploaded, so the largest files are the ones that benefit most from local processing.

Try it: Compress PDF Try it: Grayscale PDF

Choosing a target

• Email attachment limit (10–25 MB): start at quality 80 with images capped at 1800 px; go down to quality 70 and 1200 px only if it is still too large.
• Upload portals with a hard cap (2–5 MB): quality 70 and 1200 px, then grayscale if colour is not needed — a colour scan converted to grayscale is typically 30–60 % smaller. Rasterising whole pages is a last resort, because any real text becomes an image.
• Archive or legal copy: lossless only, so the document stays exactly what was signed.
• Web viewing: linearise ("Fast Web View") in addition to compressing, so the first page appears before the whole file has downloaded.
• Scanned documents you need to search: run OCR first, on the full-resolution scan, then compress by re-compressing the images. Recognition accuracy depends on the resolution and sharpness OCR sees, so downsampling first costs accuracy; the text layer adds only a few kilobytes per page and survives image re-compression (but not rasterising).

Try it: Optimize PDF for Web Try it: OCR PDF

When a PDF cannot get smaller

If a size analysis shows that images are already JPEG at a modest resolution, fonts are subset, and there are no old revisions, the file is close to its floor. Pushing the JPEG quality below 50 or the resolution below 72 dpi makes the document ugly for a few percent of savings. At that point the useful options are structural: split the document, remove pages that are not needed, or convert a scan to grayscale.

Beware of tools that always report a big saving. A compressor cannot know in advance what your file contains; one that promises "up to 90 %" on every file is describing its best case, not yours. The honest number is the one measured on your document, shown before you download.

Try it: PDF Size Analyzer Try it: Split PDF Try it: Delete PDF Pages

Frequently asked questions

References

• ISO 32000-2:2020 — Document management — Portable document format — Part 2: PDF 2.0ISO · 2020The current PDF standard: file structure, objects, pages, fonts, annotations, forms and encryption.
• PDF Reference — ISO 32000-1:2008 (PDF 1.7), Adobe copyAdobe · 2008The freely readable PDF 1.7 specification: page trees, content streams, metadata, permissions and the standard security handler.