What is PDF Compression? Deep Dive into PDF Architecture

To understand PDF compression, one must look past the visual representation and analyze the underlying ISO 32000 specification. In this technical deep dive, we explore the byte-level architecture of PDF documents and the algorithms that compress them.

The PDF Object Model

A PDF file consists of eight basic object types: Booleans, Numbers, Strings, Names (e.g. /Type), Arrays, Dictionaries, Streams, and the Null object. Indirect objects are indexed with object and generation numbers (e.g. 12 0 obj ... endobj).

Stream Objects & Compression Filters

Heavy content (text layout operators, embedded fonts, bitmap graphics) is stored in Stream Objects. Every stream object has a dictionary that specifies one or more filters applied to its binary payload:

15 0 obj
<<
  /Type /XObject
  /Subtype /Image
  /Width 1920
  /Height 1080
  /ColorSpace /DeviceRGB
  /BitsPerComponent 8
  /Filter /FlateDecode
  /Length 412890
>>
stream
... [Binary Compressed Data] ...
endstream
endobj

Object Streams & Cross-Reference Streams (PDF 1.5+)

Prior to PDF 1.5, only stream payloads could be compressed; object dictionaries, headers, and XRef tables remained uncompressed ASCII. PDF 1.5 introduced Object Streams (/ObjStm) and XRef Streams, allowing non-stream objects and cross-reference tables themselves to be compressed inside FlateDecode streams, eliminating up to 40% of overhead in vector documents.