Skip to content

Comment on Free Lossless Image Format Publications

Comments

How does it compare to fax Group 4 compression [1] or JBIG2 [2] for 1bpp -e.g. black and white- images?

[1] https://en.wikipedia.org/wiki/Group_4_compression

[2] https://en.wikipedia.org/wiki/JBIG2

Here are some results for a few black & white images from https://commons.wikimedia.org/wiki/Category:PD-ScottForesman : (I only compare to JBIG2, not Group 4, since JBIG2 is supposed to be the best of those two)

  JBIG2     FLIF      PNG
  11690    10903    27093  1971_San_Jose_Scale_Scales_on_Branch
  10458     9761    23466  1971_San_Jose_Scale_Winged_Adult
  44809    44301   105555  Accommodation_ladder
  15966    14882    44073  Aconite
  27031    25877    59353  Adder
  13569    13302    40317  Adenoids
  36409    35058    84826  African_elephant
   6890     6899    12982  Aft
  24465    23170    51297  Agate_A-40
  18969    17896    24261  Aisle
   8612     8590    23408  Allen_wrench_and_screw
   9968     9104    11792  Alpaca
  12258    11952    36657  Alternately
 196083   192611   254895  Amazon_River
  76572    73529   105637  Arabian_Sea
 275124   256820   311448  Archery_colored
 139482   137329   162432  Argonne_Forest
  40681    39952    70999  Astrolabe
  24950    23317    32007  Brace
  43808    41113   116671  Deer_2

Thank you! It looks amazing. So unless processing speed or memory requirements being a handicap, the FLIF could be a real generic graphics compression candidate :-)

Yes, that's the goal :)

Very good question. Do you or anyone else have a good test set of such images?

My gut feeling says that JBIG2 probably outperforms FLIF on repetitive images (e.g. text), since it specifically detects such patterns. On non-repetitive images, maybe FLIF is better. But it would be nice to have some benchmarks.

Patents for JBIG2 are owned by IBM and Mitsubishi.

I was interested in how they compare in compression terms. Anyway, remaining JBIG2 patents expire in 2019.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.