HotPDF Docs
Home Contents Overview

JBIG2 Compression Support

Overview

HotPDF now includes comprehensive JBIG2 (Joint Bi-level Image Experts Group 2) compression support JBIG2 is a lossy or lossless compression standard specifically designed for bi-level (monochrome) images, commonly used in PDF documents for scanned text pages and line art.

Key Features

  • High-efficiency compression for bilevel images
  • Support for both lossy and lossless compression modes
  • Support for both lossy and lossless compression modes
  • Optimised for text and line art compression
  • Significant file size reduction compared to traditional compression methods
  • Full PDF/A compliance support
  • Multi-page document support with shared dictionaries

Compression Benefits

  • Cross-page reuse: Repeated symbols are classified once and kept in shared global data
  • Symbol matching: With UseSymbolDictionary, repeated black connected components are stored once as symbols, and LossyLevel lets near-identical components share one symbol
  • Memory Efficient: Optimised for processing large scanned documents

Typical Use Cases

  • Scanned text documents
  • Technical drawings and blueprints
  • Forms and invoices
  • Maps and diagrams
  • Mixed content documents with text and graphics

Technical Implementation

The JBIG2 support is implemented through the HPDFJBIG2.pas unit, which provides:

  • THPDFJBIG2Decoder: Core decoder class for JBIG2 compressed data
  • Stream Processing: Efficient handling of JBIG2 data streams
  • Global Dictionary Support: Shared symbol dictionaries for multi-page documents
  • Scanline Access: Row-by-row data access for memory-efficient processing
  • Global Dictionary Support: Shared and multi-level chained symbol dictionaries with one-, two-, or four-byte segment identifiers
  • Cross-page batch encoder: THPDFJBIG2Encoder.EncodePagesToByteArrays trains across the complete batch and returns one page stream per input plus shared global data
  • Batch contract: TJBIG2BatchPage describes one packed page, TJBIG2BatchEncodeBackend identifies the selected implementation, and TJBIG2BatchEncodeCallback defines external batch integrations
  • Document integration: THotPDF.AddJBIG2ImageBatch and THotPDF.AddJBIG2BitmapBatch encode before mutating the PDF graph, attach a shared /JBIG2Globals stream, and return image indexes in input order
  • Batch telemetry: TJBIG2BatchEncodeStatistics reports selected backend, input and output sizes, native symbol count, and budget rejection
  • Backend recovery: HPDFJBIG2EncRegisterBackends restores both built-in encoder callbacks after temporary application overrides are cleared
  • HPDFJBIG2SplitSymbolDictionaryGlobals: Separates a sequential embedded stream into global dictionaries and page image segments while preserving long-form dependency headers
  • Visible-Region Decode: THPDFJBIG2Decoder.LoadRegionFromByteArray allocates only the requested window, resolves its transitive dictionary dependencies, composes intersecting page regions with clipping, and skips independent off-window regions before entropy decoding
  • Region Decode Telemetry: LastRegionDecodeStats reports decoded and skipped region counts plus decoded source pixels, while UsedIncrementalRegionDecode distinguishes the native window path from complete-decode fallback
  • Scanline Access: Row-by-row data access for memory-efficient processing

Usage Example


// Delphi example - Processing JBIG2 compressed data
procedure ProcessJBIG2Data;
var
  PDF: THotPDF;
  Decoder: THPDFJBIG2Decoder;
  ImageData: TBytesArray;
  ScanlineData: TBytesArray;
  Row: Integer;
begin
  PDF := THotPDF.Create(nil);
  Decoder := THPDFJBIG2Decoder.Create;
  try
    PDF.BeginDoc;
    PDF.AddPage;

    // Load JBIG2 compressed image data
    if Decoder.LoadFromByteArray(ImageData) then
    begin
      // Process scanline by scanline for memory efficiency
      for Row := 0 to Decoder.Height - 1 do
      begin
        if Decoder.GetScanline(Row, ScanlineData) then
        begin
          // Process scanline data as needed
          // Add to PDF page
        end;
      end;
    end;

    PDF.EndDoc;
  finally
    Decoder.Free;
    PDF.Free;
  end;
end;

Encoder Options and Backends

  • THotPDF.JBIG2Options: The document's THPDFJBIG2Options, which AddImage(..., icJBIG2) converts into TJBIG2EncodeOptions. Its defaults suit scans: Lossless=True, LossyLevel=5, RemoveIsolatedBlackPixels=True and BlackDotSize=3
  • Speckle removal: Whilst RemoveIsolatedBlackPixels is True, black connected components whose width and height are both smaller than BlackDotSize are cleared before encoding, so the defaults remove dots of 2 x 2 pixels and smaller. Switch it off when the output must match the bitmap pixel for pixel
  • Encoder choice: UseExternalEncoder (default True) uses the registered high-performance arithmetic encoder; when it is off, missing or fails, the built-in lossless MMR generic region is written instead. UseNativeArithmeticFallback=True tries the built-in Pascal arithmetic coder before MMR, and with UseSymbolDictionary=True it first attempts a symbol dictionary with text regions
  • DefaultJBIG2EncodeOptions: The options a stand-alone THPDFJBIG2Encoder uses: lossless, with LossyLevel=0 and no speckle removal
  • Decoder choice: THotPDF.UseExternalJBIG2Decoder (default True) decodes loaded /JBIG2Decode images with the high-performance backend where one is available; False uses the built-in Pascal decoder
  • Custom backends: RegisterJBIG2EncoderBackend and RegisterJBIG2DecoderBackend install application callbacks, IsJBIG2EncoderBackendAvailable and IsJBIG2DecoderBackendAvailable report them, and ClearJBIG2Backends removes them again

Class Reference

THPDFJBIG2Decoder

  • LoadFromByteArray: Load JBIG2 data from byte array; gives False when the data is incomplete or not supported
  • GetScanline: Retrieve specific scanline data
  • Width/Height: Image dimensions properties

Performance Optimisation

  • A fixed 100-page scan corpus compresses from 158,300 bytes with separate generic regions to 33,600 bytes with external cross-page training, a 78.8 per cent reduction
  • The same corpus completes in 11 ms instead of 53 ms on Win64 and 16 ms instead of 77 ms on Win32 in the regression benchmark
  • Page-to-symbol lookup is indexed and repeated encoded page streams are cached for constant-time batch extraction
  • Memory-efficient scanline processing
  • Optimised for large document handling
  • Minimal memory footprint during decompression

Standards Compliance

  • ITU-T T.88 standard compliance
  • ISO/IEC 14492 standard support
  • PDF 1.4 or later image XObjects with /JBIG2Decode and /JBIG2Globals
  • HotPDF does not write JBIG2 whilst PDF/A output is active: AddImage(..., icJBIG2) writes icCCITT42 instead and batch insertion gives False
  • Cross-platform consistency

See Also