Compress PDFs by knowing where the size actually is

Diagnosing the file

Where the bytes in a PDF actually go

"Compressing a PDF" is a vague action because PDFs aren't a single thing. A 30 MB document gets to 30 MB through some combination of high-DPI embedded images, fully-embedded font glyph tables, accumulated unused objects from prior revisions, redundant resource dictionaries, and uncompressed content streams. Optimization that doesn't know which of those surfaces is doing the damage produces unimpressive results — you save 5%, you keep the bloat.

PDF Batch Editor's optimizer addresses each surface independently. The next four sections walk through what each technique does and when it pays off. The compression profiles (Web, Balanced, Print, Custom) are presets that combine these techniques at different intensities; the Custom profile lets you control each independently.

Technique 1

Image downsampling and recompression

In most documents that aren't pure text, embedded raster images dominate the file size. A scan saved at 600 DPI is twice as many pixels per inch in each dimension as a 300 DPI scan — four times the data, for an output that displays the same on screen and prints indistinguishably from 300 DPI on most office printers. Downsampling computes the effective DPI of each embedded image relative to its rendered size on the page, and resamples it to a target DPI when the original is higher.

Targets per profile:

Web        96 DPI    Aggressive — for screen viewing and email
Balanced   150 DPI   Sweet spot — readable on screen, prints well at small sizes
Print      300 DPI   Conservative — print-shop quality
Custom     any       Set per-image-type and per-format

After downsampling, images are recompressed with format-appropriate codecs — JPEG for photographic content, lossless (Flate or JBIG2) for line art and screenshots. The Custom profile lets you choose codec per image type, set quality factors for JPEG independently, and exclude image classes (e.g. don't downsample any image used as a logo) from the pass.

Technique 2

Font subsetting to the glyphs actually used

When a PDF embeds a font, it usually embeds the whole thing — every glyph, every weight, every style. A standard Latin font has roughly 250 glyphs at one weight. A Pan-European font has 800. A CJK font has tens of thousands. Embedding the full glyph table for a font your document barely uses is dead weight on every page reference.

Subsetting analyzes the content streams to find which glyphs are actually rendered, and writes a new font program containing only those. A document that uses one weight of one font with 200 unique characters can drop a 400 KB embedded font down to 30 KB. Across a 200-document batch with three or four fonts each, that is meaningful aggregate savings — and it changes nothing about how the document looks.

Technique 3

Unused objects, duplicate resources, and stream recompression

Every time a PDF is edited and saved with an incremental update, the previous revision's objects stay in the file — they are simply no longer referenced. A document that's been through ten revisions accumulates ten generations of unused content. The optimizer rebuilds the cross-reference table without the dead objects.

Resources can also duplicate. A logo embedded into 50 forms by a poorly-implemented form generator may end up as 50 copies of the same image. The optimizer detects byte-identical resources and consolidates references to a single copy.

Finally, content streams that were saved uncompressed (or with weak compression) are re-compressed using FlateDecode at a higher level. Object streams group related dictionary objects into a single compressed stream, reducing per-object overhead. None of these techniques touch what the document looks like.

Verify the work

Before/after measurement and the operation log

Optimization without measurement is hand-waving. Every batch produces a per-file before/after report: original size, optimized size, percentage reduction, and the dominant technique that saved the most bytes (so you know whether the win came from images, fonts, or structural cleanup). The total batch savings are summed at the bottom — "200 files, 4.7 GB → 1.9 GB, 60% reduction."

Files that were already efficient (a 200 KB text-only PDF with subset fonts and no images) pass through with a near-zero change instead of being touched unnecessarily. The operation log records every file processed so the audit trail matches the actual work done.