JBIG2 Encoder

This is an encoder for JBIG2.

JBIG2 encodes bi-level (1 bpp) images using a number of clever tricks to get better compression than G4. This encoder can:

• Generate JBIG2 files, or fragments for embedding in PDFs
• Generic region encoding
• Perform symbol extraction, classification and text region coding
• Perform refinement coding and,
• Compress multipage documents

It uses the (Apache-ish licensed) Leptonica library. Version 1.74 or later is required.

The refinement coding causes Acrobat to crash. It's not known if this is a bug in Acrobat, though it may well be.

Note: Windows Command Prompt does not support wildcard expansion, so *.jpg will not work. You'll need to manually expand the file names yourself or you need to use the latest git code and MSVC build.

See the jbig2enc.h header for the high level API, or the jbig2 program for an example of usage:

jbig2 -s -a -p -v -T 100 -b imgs images/*.jpg && python jbig2topdf.py imgs >out.pdf

or with standalone mode:

jbig2 -a -p -v images/feyn.tif > feyn.jb2 && python jbig2topdf.py -s feyn.jb2 > feyn.pdf

to encode jbig2 files for pdf creation. If you want to encode an image and then view output first to include in pdf

jbig2 -s -S -p -v -O out.png images/*.jpg

If you want to encode an image as jbig2 (can be view in STDU Viewer on Windows) run:

jbig2 -s images/feyn.tif >feyn.jb2

In symbol mode the classifier assigns every component instance to a symbol class and records where the class template is placed. That information is consumed into the arithmetic-coded text regions and normally lost to callers. With --emit-json (requires -s) it is written to a JSON sidecar, so the encoder can double as a glyph clustering layer for downstream tools:

jbig2 -s -p -b output --emit-json output.json images/amoris.2.150.jpg

This writes the usual symbol mode outputs (output.sym, output.0000) plus output.json with the provenance of every instance:

{
  "version": 1,
  "num_pages": 1,
  "num_symbols": 929,
  "num_instances": 3089,
  "pages": [
    {"page": 1, "width": 855, "height": 1263, "xres": 150, "yres": 150}
  ],
  "symbols": [
    {"class": 0, "width": 18, "height": 264}
  ],
  "instances": [
    {"class": 0, "page": 1, "ul": [-1, -1], "ll": [0, 263]}
  ]
}

All coordinates are raster coordinates of the page at encode resolution: x is measured from the left edge, y from the top edge. For each instance, ul is the upper left corner of the template placement box and ll the lower left corner of the ink, which is the placement used by the JBIG2 text region coder. Both corners come from the classifier's placement arrays, and for a component touching the top or left crop edge the anchor can fall a pixel outside the page raster, so small negative values as in the sample are expected and not an error. symbols holds one entry per class with the exemplar dimensions in pixels. Pages that were not encoded in symbol mode report a zero instance count. The flag is default-off and writes only to the named file, leaving the stdout/stderr contract unchanged.

• jbig2enc-samples
• jbig2enc-minidjvu