Re: New lossless image compression algorithm

Stefano Brocchi wrote:

I have developed a new lossless compression algorithm for true color
images, and created a website that describes it in detail. The
obtained compression ratios are still not comparable to the state of
art, but benchmarks report that the new algorithm outperforms the
widely used format PNG and gives results comparable to those of the
lossless mode of Jpeg2000.

State of the art lossless compression is actually JPEG-LS, not JPEG2000,
which has a low-complexity ultra-fast compression technique. It typically outperforms JPEG2000. The entropy coding technique used here is Golomb-coding (as simple as it is!).

The algorithm is based on Huffman coding to keep time requirements
low. There are also two new techniques (to my knowledge) that are
worth mentioning: a color filtering phase, made to reduce color
correlation on a local basis, and the use of polyominoes, to group and
encode conveniently similar values close to each other in the image.

On the site both the program and the Java source code are available
for download. There are also some pages about benchmark results and an
explanation of how the algorithm works. The link to the site is

http://www.researchandtechnology.net/pcif/index.php

The work on the algorithm is actually under development, so
suggestions are welcome

A couple of comments, maybe: The waterloo corpus is aged, and the images you use are too small to be useful. Niels Fröhlich recently made an extended corpus available, and the corpus used by the JPEG is also available (but beware, this is *huge*. Several hundred MBs, acutally.)

Concerning JPEG2000, this was mostly optimized towards photographic images and not computer generated images. If you want to compress the latter, JBIG2 might be your best bet (and not even JPEG-LS, which is also designed with natural images in mind). There is a recent effort to apply it to color images.

One of the design goals of JPEG2000 was to offer ultimate flexibility, which is why certain coding options haven't been considered. One of them is a better inter-color decorrelation, another the usage of inter-band correlations (e.g. the SPIHT algorithm uses them with good success, but wasn't picked because it's not as scalable as the EBCOT was).

You may find it amusing that in part-2 of JPEG2000 the color-transformation can be defined freely, and can be re-defined from tile to tile, i.e. each image tile can select from a set of pre-defined transforms. I'm not aware of a codec that really tries to exploit this, but there are at least two codecs out that would be able to decode those codestreams correctly. (-:

I would furthermore consider a more modern entropy coding backend, i.e. arithmetic coding.

Concerning Polynomios, I'm unclear whether this is a good model for image data. It would be helpful to give some insight into this.

So long,
Thomas


.

Relevant Pages

  • Re: Differential intra-interchannel filters for RGB images lossless compression
    ... >>> However IMHO many photographic images are even slightly more difficult ... don't look right, lighting doesn't look right, surfaces are almost always ... compression, it seems concievable to me that they could. ... would likely be able to fairly accurately predict the pixel. ...
    (comp.compression)
  • Re: New lossless image compression algorithm
    ... Well the algorithm should be quite stable, ... "Lossless Compression and Interpolation for High Quality Still Images ... whereas LOCO allready is in cameras ...
    (comp.compression)
  • Re: How much lossless compression is possible in images?
    ... >So how much compression could you get? ... compression that halts early, given that the ideal algorithm for that ... on half of the images (99.99999999999% of 4GB images ... the data will just be a UTM copy algorithm with the 4GB input string ...
    (sci.math)
  • Re: Microsoft HDPhoto algorithm exposed
    ... how a new image file format would be really 'new'? ... I'm not clear whether compression factors really matter too much. ... JPEG 2000 probably got like 10-20% better at ten times the complexity, JPEG-XR didn't get much better at about twice to three times the complexity. ... A considerably useful feature would be to provide a format that would allow me to store "relatively unprocessed" images, ...
    (comp.compression)
  • Re: The future (or lack thereof?) of CDs
    ... audio with regard to compression, ... produce images to match the quality of large format images produced by ... generation seems to witness this compaction: recently from 4x5 inch ...
    (rec.music.classical.recordings)
Loading