PDF Compression Explained: How It Works and Best Practices

PDF is the de facto standard for sharing documents — from contracts and invoices to research papers and product manuals. Yet a single PDF can range from a few kilobytes to hundreds of megabytes depending on its content and how it was created. When files are bloated, real-world problems follow: email attachments bounce against size limits, web pages load slowly, cloud storage costs creep upward, and mobile users burn through data plans downloading files they could have received at a fraction of the size.

File size also has direct implications for search engine optimization and user experience. Google considers page speed a ranking factor, and PDFs embedded or linked from web pages contribute to overall load time. For businesses that host downloadable catalogs, whitepapers, or legal documents, compressing those PDFs is not just a convenience — it is a performance requirement.

Understanding how PDF compression works gives you the knowledge to shrink files intelligently rather than blindly re-saving and hoping for the best. In this article, we will walk through the internal structure of a PDF, the compression algorithms that operate on different content types, and the best practices professionals use to balance quality against file size. If you want to jump straight to compressing a file, try our Compress PDF tool — but read on if you want to understand what happens under the hood.

A PDF file is not a flat image or a simple text stream. It is a structured binary format defined by the ISO 32000 family of standards (the latest being PDF 2.0 — ISO 32000-2, published in 2020). At its core, a PDF is a collection of numbered objects arranged in a specific layout. Understanding this layout is the first step toward understanding where compression opportunities exist.

Every PDF consists of four major sections. The header declares the PDF version (e.g., %PDF-1.7 or %PDF-2.0). The body contains all the objects that make up the document — page descriptions, fonts, images, metadata, annotations, and more. Each object has a unique identifier (an object number and a generation number) and can be of several types: booleans, numbers, strings, names, arrays, dictionaries, and streams. Streams are the workhorses of PDF; they hold the actual binary data for images, embedded fonts, and page content instructions.

After the body comes the cross-reference table (often abbreviated as xref). This table maps each object number to its byte offset within the file, allowing PDF readers to jump directly to any object without scanning the entire file sequentially. Finally, the trailer provides the byte offset of the xref table itself plus pointers to the document catalog (the root object of the page tree) and optional encryption dictionaries.

Why does this matter for compression? Because each of these sections — especially the streams in the body — can be compressed independently using different filters. A single PDF might contain a JPEG-compressed photograph, a Flate-compressed vector drawing, and an uncompressed cross-reference table all in the same file. Optimizing a PDF means choosing the right compression strategy for each object type.

In the vast majority of PDFs, images account for 80-95% of the total file size. A single uncompressed 300 DPI photograph on a letter-size page occupies roughly 25 megabytes of raw pixel data. Clearly, image compression is where the most dramatic file-size reductions occur.

PDF supports several image compression filters, each suited to different image characteristics. DCTDecode applies JPEG compression — a lossy algorithm that exploits the human visual system's lower sensitivity to high-frequency color detail. JPEG works exceptionally well for photographs and complex color images, achieving compression ratios of 10:1 to 20:1 with minimal perceptible quality loss. When you use our JPG to PDF converter, the resulting PDF embeds the JPEG data directly via DCTDecode.

FlateDecode (based on the zlib/deflate algorithm) provides lossless compression. It is the default filter for most non-image streams in a PDF but is also used for images where lossless fidelity is required — such as screenshots, diagrams, and charts with sharp text. Flate typically achieves 2:1 to 5:1 compression on image data, far less than JPEG, but preserves every pixel exactly.

For scanned black-and-white documents, two specialized filters exist. CCITTFaxDecode implements the CCITT Group 3 and Group 4 fax compression algorithms, designed specifically for bilevel (1-bit) images. Group 4 is particularly efficient, often compressing scanned text pages to 5-10% of their uncompressed size. JBIG2Decode goes further: it uses pattern matching and symbol dictionaries to identify repeated glyphs (like the letter "e" appearing hundreds of times on a page) and stores each unique symbol once. JBIG2 can achieve compression ratios of 50:1 or higher on text-heavy scanned pages, making it the gold standard for document imaging systems.

Beyond choosing the right codec, image downsampling — reducing the pixel resolution of images — is another powerful lever. A photograph scanned at 600 DPI for archival purposes only needs 150 DPI for on-screen viewing and 300 DPI for high-quality printing. Downsampling algorithms determine how pixel values are recalculated when the resolution decreases. Bicubic interpolation produces the smoothest results by considering the 16 nearest pixels, making it ideal for photographs. Bilinear interpolation uses 4 neighboring pixels and is faster but slightly less smooth. Nearest-neighbor (subsampling) is the fastest but can produce jagged edges — acceptable only for bilevel images. Most professional PDF optimizers default to bicubic downsampling, and our Compress PDF tool applies similar techniques to reduce image resolution intelligently.

Fonts are the second-largest contributor to PDF file size after images. A full OpenType font file can weigh 1-5 MB, and a PDF that embeds four or five complete fonts can easily add 10-20 MB of overhead — even if the document only uses a few dozen characters from each font.

Font subsetting solves this by embedding only the glyphs (character shapes) that actually appear in the document. If your PDF uses the word "Hello" in a particular font, subsetting includes only the outlines for H, e, l, and o — discarding the other thousands of glyphs. This can reduce a 2 MB font to 20-50 KB. Most modern PDF creation tools perform subsetting automatically, but documents generated by older software or certain enterprise workflows may still embed full fonts unnecessarily.

The opposite of subsetting is full font embedding, where the entire font program is included. This is required in certain archival scenarios — notably the PDF/A standard (ISO 19005) — to ensure the document can be rendered identically on any system regardless of installed fonts. PDF/A-1b requires all fonts to be embedded (subsetting is permitted), while PDF/A-2 and PDF/A-3 relax some constraints. The tradeoff is clear: full embedding guarantees fidelity but increases file size.

Another optimization is font deduplication. When merging multiple PDFs (for example, using a Merge PDF tool), the same font may be embedded independently in each source file. A smart optimizer detects duplicate font programs and keeps only one copy, updating all references accordingly. This alone can save megabytes in merged documents.

Finally, fonts embedded as Type 1 (PostScript) outlines are generally larger than their CFF (Compact Font Format) or TrueType equivalents. Converting Type 1 fonts to CFF during optimization can shave 20-40% off the font data size without any visual difference, since CFF uses a more compact encoding for glyph outlines.

Every page in a PDF has a content stream — a sequence of operators that describe where to place text, how to draw lines, and how to render colors. These streams are plain text by default and often highly compressible because they contain repetitive operator sequences like BT (begin text), Tf (set font), Td (move text position), and Tj (show string).

The standard compression filter for content streams is FlateDecode, which wraps the widely-used zlib library implementing the DEFLATE algorithm (RFC 1951). DEFLATE combines LZ77 sliding-window matching with Huffman coding to compress data losslessly. On typical page content streams, Flate achieves compression ratios of 3:1 to 8:1 depending on the complexity of the page layout.

PDF also supports filter chaining — applying multiple filters in sequence. A common pipeline is /ASCIIHexDecode followed by /FlateDecode, which first decodes a hex-encoded stream back to binary and then decompresses it. During optimization, these chains can sometimes be simplified. For example, removing an unnecessary /ASCII85Decode layer that was added for compatibility with older PostScript workflows can reduce the raw stream size before Flate even begins its work.

In PDF 1.5 and later, object streams and cross-reference streams allow the xref table and small objects to be compressed together as a single Flate-encoded stream, replacing the older plaintext xref format. This is one of the reasons that re-saving an older PDF in a modern viewer can sometimes reduce file size even without touching the images or fonts — the structural overhead itself gets compressed.

The fundamental tension in PDF compression is between file size and visual or data fidelity. Lossy techniques — JPEG recompression, image downsampling, and color space conversion — achieve dramatic size reductions but permanently discard information. Lossless techniques — Flate encoding, font subsetting, metadata stripping, and structural optimization — preserve perfect fidelity but yield more modest gains.

For photographs and full-color images, lossy compression is almost always acceptable. Recompressing a JPEG image at quality level 75 (on a 0-100 scale) reduces file size by roughly 50% compared to quality 95, with differences that are nearly imperceptible to the human eye on screen or in print. Below quality 50, however, blocking artifacts become visible and text overlaid on images may become unreadable.

For line art, technical drawings, and charts, lossless compression is strongly preferred. JPEG compression introduces ringing artifacts around sharp edges — a phenomenon known as the Gibbs effect — that can make thin lines appear blurry and graph labels illegible. These content types compress well with FlateDecode and should not be JPEG-compressed even when the rest of the document's photographs are.

For scanned documents containing primarily text, converting from full-color JPEG to bilevel (1-bit) JBIG2 or CCITT Group 4 is often the single most impactful optimization. A 10 MB color scan of a text page can shrink to under 100 KB when thresholded to black-and-white and compressed with JBIG2. The tradeoff: any faint background color, handwritten annotations in light ink, or colored logos will be lost in the conversion. If you need to extract images from a PDF first to inspect them, our PDF to JPG tool can help you preview each page as a raster image before deciding on a compression strategy.

A balanced approach for most documents is to apply lossy compression to photographic images while keeping vector graphics and text streams lossless. This hybrid strategy, which professional tools call mixed-mode compression, typically achieves the best ratio of size reduction to quality preservation.

Beyond compressing content, the arrangement of data within a PDF file affects both perceived performance and actual file size. Linearization (sometimes called "fast web view") reorganizes a PDF so that the first page's data appears at the beginning of the file. This allows PDF viewers — especially web-based ones — to display the first page while the rest of the file is still downloading. Linearization does not reduce file size, but it dramatically improves the user experience for large documents served over the web.

Object deduplication scans the document for identical objects (same dictionary contents, same stream data) and merges them into a single shared object. This is common after merging multiple PDFs or after copy-pasting content between pages. Deduplication can remove several percent of the total file size without any quality impact.

Removing unused objects is another structural optimization. PDFs are append-only by design: when you delete a page or remove an annotation, the original objects remain in the file — they are simply unlinked from the page tree. This is why a "saved" PDF may be larger than a "Save As" copy. A proper optimizer performs garbage collection, discarding unreferenced objects and rebuilding the xref table.

Stripping metadata and hidden content can also contribute to size reduction. PDFs often contain embedded thumbnails (preview images for each page at low resolution), document-level metadata (author, creation software, revision history), and XMP (Extensible Metadata Platform) packets. While individually small, these elements add up in large document sets. Metadata removal should be done thoughtfully — PDF/A compliance, for example, requires certain XMP metadata to be present.

There is no single "optimal" compression setting — the right approach depends on how the PDF will be used. Here are recommended strategies for the most common scenarios.

Email attachments (target: under 10 MB). Downsample all images to 150 DPI using bicubic interpolation. Recompress JPEG images at quality 70-75. Subset all fonts. Remove metadata, thumbnails, and unused objects. Enable cross-reference stream compression. For most office documents, this produces files well under 5 MB while maintaining good readability on screen.

Web hosting and downloads (target: fast load times). Apply all the email optimizations plus linearization for fast web view. Consider converting full-color scans to grayscale if color is not essential — this halves the image data before any codec runs. Serve PDFs with HTTP compression (gzip or Brotli) enabled at the server level for additional transport-layer savings of 10-20% on the already-compressed file.

Print production (target: maximum quality). Keep images at 300 DPI minimum (600 DPI for line art). Use lossless Flate compression for all images or, if lossy is necessary, JPEG quality 90+. Embed full fonts (subsetting is acceptable, but ensure all used glyphs are present). Do not strip color profiles (ICC profiles are critical for accurate color reproduction on press). Linearization and metadata stripping are irrelevant here — fidelity is paramount.

Long-term archival — PDF/A (ISO 19005). The PDF/A standard restricts which compression features are allowed. JBIG2 compression is prohibited in PDF/A-1 because certain JBIG2 patent holders created uncertainty about its long-term accessibility (PDF/A-2 allows it). All fonts must be fully embedded. JPEG 2000 (JPXDecode) is allowed in PDF/A-2 and later, offering better quality-per-byte than traditional JPEG at higher compression ratios. Transparency, JavaScript, and encryption are all forbidden. When creating archival PDFs, validate against the specific PDF/A conformance level (1a, 1b, 2a, 2b, 2u, 3a, 3b) using a dedicated validator.

Scanned document workflows. Apply automatic deskewing and despeckling before compression. Convert text regions to bilevel and compress with CCITT Group 4 or JBIG2. Keep photographic regions (embedded pictures, letterheads with gradients) as separate JPEG-compressed image objects. Apply OCR to create a hidden text layer for searchability — this adds a modest amount of data but makes the PDF far more useful. The combination of aggressive bilevel compression plus an OCR text layer often produces files that are smaller than the original color scans while being fully searchable.

The landscape of PDF compression tools ranges from command-line utilities to cloud services. Understanding what each does under the hood helps you choose the right one.

Ghostscript is the open-source workhorse. Its -dPDFSETTINGS flag offers presets like /screen (72 DPI, aggressive JPEG), /ebook (150 DPI, medium JPEG), /printer (300 DPI, high-quality JPEG), and /prepress (300 DPI, minimal lossy compression). Ghostscript re-renders the entire PDF through its PostScript interpreter, which can fix broken PDFs but may also subtly alter fonts or colors. It is powerful but demands careful parameter tuning.

QPDF focuses on structural optimization: linearization, object stream compression, unreferenced object removal, and xref stream conversion. It does not recompress images or subset fonts, making it a perfect complement to Ghostscript — run QPDF after Ghostscript for a clean, well-structured output.

Commercial tools like Adobe Acrobat Pro's "Save As Optimized PDF" dialog expose granular controls over image resampling thresholds, font embedding policy, and transparency flattening. Adobe's optimizer is well-tested across edge cases but requires a paid license.

Browser-based tools like our Compress PDF provide the most accessible experience. You upload a file, choose a compression level, and download the optimized result — no software installation required. Because processing happens entirely in the browser using WebAssembly and JavaScript libraries, your files never leave your device, preserving privacy. The tradeoff is that browser tools may not support every advanced optimization (e.g., JBIG2 encoding) that a desktop tool can perform, but for the vast majority of everyday documents they produce excellent results.

For batch workflows involving hundreds or thousands of PDFs, server-side pipelines using Ghostscript, QPDF, and custom scripts are the most efficient approach. For one-off documents or quick share-ready compression, a browser-based tool eliminates friction and gets the job done in seconds.

PDF compression is not a single technique — it is a layered process that targets images, fonts, content streams, and document structure independently. The most effective approach combines lossy image compression for photographs, lossless encoding for vector graphics and text, font subsetting to eliminate unused glyphs, and structural optimization to remove dead weight.

Always start by identifying what is making a PDF large. If images dominate, focus on downsampling and recompression. If fonts are the culprit, verify that subsetting is enabled and duplicates are merged. If the file has been edited many times, a simple "Save As" to a new file can discard orphaned objects and reclaim significant space.

Remember the standards: PDF/A (ISO 19005) for long-term archival, PDF 2.0 (ISO 32000-2) for the latest feature set including modern encryption and richer metadata. When compliance with these standards matters, choose your compression settings accordingly and validate the output.

Whether you are preparing a contract to email, optimizing a product catalog for your website, or archiving invoices for regulatory compliance, the principles in this guide apply. Use our Compress PDF tool for a quick, privacy-respecting compression right in your browser, or combine it with Merge PDF and JPG to PDF for complete PDF workflows — no uploads to external servers required.