PDF Compression Explained: Reduce File Size Without Losing Quality

What makes a PDF file big in the first place?

A PDF is a container, and its size comes from whatever you put in it. Text is nearly free: a thousand pages of plain text often weigh less than a single megabyte. The weight comes from four things:

• Embedded images. A full-page photo at 300 DPI can be several megabytes. Multiply by forty pages and you have a file nobody can email.
• Embedded fonts. A full font family with all weights can add megabytes. Most documents use a fraction of one weight.
• Unused and duplicate objects. Design apps export layers, hidden elements, and alternate versions. The same logo image may be stored separately on every page instead of once.
• Edit history. PDF editors save incrementally by appending changes, so every round of edits adds dead weight.

Understanding this list is the whole game. Compression is just the systematic removal of each of these, starting with the biggest.

How does image downsampling work?

Downsampling is the single most powerful compression technique, because images are the single biggest contributor to file size. It works like this: an image stored at 300 dots per inch contains four times as many pixels as the same image at 150 DPI (half the resolution in each direction means one quarter of the pixels). If nobody will ever print the document at high resolution, those extra pixels are pure waste.

The practical numbers: 72 to 150 DPI is plenty for on-screen reading. 150 DPI looks clean on an office laser printer. 300 DPI is for professional print work. Anything above 300 in a PDF is almost never justified.

My recommendation: downsample color photos to 150 DPI with bicubic resampling for general documents. Bicubic averages neighboring pixels smoothly, so the result looks natural rather than jagged. Reserve 300 DPI for files headed to a print shop, and check the result at 200 percent zoom before you send it anywhere important.

What is font subsetting?

When a PDF embeds a font, it can embed the entire font file, every letter, number, symbol, and glyph in every style, even if the document only uses basic Latin text. Font subsetting embeds only the characters the document actually uses. A document set in regular body text might need fewer than a hundred glyphs out of the thousands in a full font file.

There is a related trick: the fourteen standard PDF fonts (Times, Helvetica, Courier, and their variants) do not need to be embedded at all. Every PDF reader is required to have them. Stripping embedded copies of standard fonts and replacing them with references is pure savings with zero visual change.

The honest caveat: subset fonts can cause problems if someone later edits the PDF and types a character that was not in the original subset, like adding an accented letter to a document that only contained plain ASCII. For final documents nobody will edit, subset freely. For templates people will fill in, keep the full font.

What are object streams and why do they matter?

Inside every PDF, the page content, the text positioning commands, the vector drawings, is stored as data objects. By default these objects are stored uncompressed, readable as plain text if you open the PDF in a text editor. Object stream compression runs them through Flate (the same deflate algorithm behind ZIP files), typically cutting that portion of the file to a third or less of its original size.

This is entirely lossless. The text, the layout, the vectors come back bit-for-bit identical when the reader decompresses them. The only cost is a tiny amount of CPU time when opening the file, which no modern device will notice.

Optimizers also remove unused objects at this stage: orphaned resources from deleted pages, duplicate images stored per page, and metadata nobody reads. A document assembled from many sources, like a merged report, often carries a surprising amount of this dead weight.

Lossy vs lossless: which should you pick?

Lossless methods cost you nothing visible. Always apply them: object stream compression, font subsetting, metadata removal, unused object cleanup. There is no reason not to.

Lossy methods trade pixels for bytes, and the right choice depends on the image type:

• Photographs: JPEG compression. It is designed for continuous-tone images and gives the best size-to-quality ratio for photos. A quality setting around 70 to 80 percent is visually transparent for most uses.
• Black-and-white scans: CCITT Group 4, the fax-machine standard. It compresses bi-level images (pure black and white, no gray) extremely well, far better than JPEG, which is built for color gradients.
• Line art and diagrams: Flate (ZIP) compression. Lossless, and ideal for flat colors and sharp edges where JPEG would smear.
• Color scans of text pages: convert to grayscale or black and white first. A color scan of a black-text page is paying full color price for zero color benefit.

Here is the judgment call I make on every file: lossy for images only, lossless for everything else, and never recompress a JPEG image twice. Each JPEG pass throws away more detail, and the damage accumulates.

Why does saving sometimes make a PDF bigger?

Because of incremental updates. When you edit a PDF in most readers and hit save, the software appends your changes to the end of the file and marks the old content as superseded. The old content stays in the file. Edit a document ten times and you are carrying ten generations of changes.

The fix is simple: use "Save As" instead of "Save," or run an explicit optimize pass. Both rewrite the file cleanly and drop the superseded data. If a file has been edited many times, a clean rewrite alone can cut its size in half with no other changes.

When should you NOT compress a PDF?

Compression is not always the right move. Skip it, or keep an uncompressed master, when:

• Print production files. A file headed to a commercial printer should stay at full resolution with full fonts. Let the print shop handle optimization.
• Archival and legal originals. Courts, regulators, and archives may require the original fidelity. Compress a copy for distribution, keep the original untouched.
• Documents with fine detail. Maps, engineering drawings, and medical images can lose critical information to downsampling. What looks fine at screen zoom can fail under scrutiny.
• Files you will keep editing. Compress at the end of the workflow, not the middle. Every lossy pass degrades the source for the next edit.

How do you compress a PDF safely, step by step?

Follow this order and you will never damage a file:

1. Back up the original. Save a copy before you touch anything. This is the step everyone skips and regrets.
2. Apply lossless optimization first. Object streams, font subsetting, metadata and unused object removal. Check the file still looks right.
3. Downsample images to your target. 150 DPI for screen and office printing, 300 DPI for professional print.
4. Verify visually. Open the compressed file, zoom to 200 percent, and check small text, thin lines, and photo detail against the original.
5. Test functionality. Confirm links, bookmarks, form fields, and text selection all still work. If anything broke, the tool rebuilt the file instead of optimizing it.
6. Compare sizes. If the savings are under 10 percent, the file was already lean. Ship it and move on.

What does a good compression tool do under the hood?

A serious optimizer works through the file in a fixed order, biggest savings first. It starts by inventorying every image: resolution, color depth, and compression method. Photos get downsampled to your target DPI and recompressed as JPEG. Black-and-white pages get converted to bi-level and compressed with a fax-style method. Then it subsets fonts, compresses object streams, removes duplicate resources, strips metadata and thumbnails, and finishes with a clean rewrite that drops all the incremental edit history.

You can judge any tool by what it lets you control. If it offers per-image-type settings (color, grayscale, monochrome separately), font subsetting options, and a choice of what metadata to strip, it is a real optimizer. If it offers one button and no settings, it is guessing, and guesses sometimes damage files. The best tools also show you a before-and-after breakdown by category, so you can see that 80 percent of the savings came from images and decide whether the result is worth it.