Hi folks,

just had a nice chat with the organizers of this year's PCS (Picture Coding Symposium, see http://www.pcs2015.org), amongst them David Taubman. David is the creator of the JPEG 2000 standard, in case you do not know him. We would like to organize an "Image Compression Challenge" at this year's PCS, in addition to the prediction challenge that is already there.

Thus, we are looking for contributions, in the form of codecs and/or images to test on, and/or objective metrics. The challenge will include subjective testing (i.e. by human observers) and objective testing (computer based algorithms). It's, however, not only raw coding performance we're looking at, but also "features", like scalability, extraction of regions, regions of interest, privacy protection, complexity vs. performance and so on.

The rules aren't fixed yet, and you've here the chance to discuss and refine them, in addition to providing codecs.

How to contribute:
- ideally, source code. If you need an NDA, that could be organized.
- failing that, a linux binary
- failing that, a windows binary (must run under wine, non-interactively)

The codec must accept an sRGB image in the PPM binary format (big endian), and a parameter to control a "quality" (bitrate, quality, quantization...) and generate a stream in its own format. A decoder must be provided that generates a PPM image as output.

- Input and output images must be in PPM format, 8 bit, and - if possible - 12 bit.
- a single parameter to control the bitrate, all other parameters will be fixed and will follow the suggestions of the authors
- must be completely self-running, no user interaction possible

In addition to the image set we already have, you may provide additional images for testing. These images must be in sRGB or ITU T.601 (8 bit or 12 bit), i.e. gamma corrected. We might have to scale and/or crop the images to fit on the monitor for subjective testing, and will then measure (objecitvely/subjectively) on these cropped/scaled images.

- The current test image set consists of the JPEG AIC core1 and core2 test images, the latter downscaled to 12-bit resolution.

The test protocol for subjective evaluation is not yet fixed, likely a side-by-side comparison with a five-degree impairment scale rating.
As objective metrics, we will at least include (this is an open list):

- peak error
- multiscale SSIM
- running time
- memory footprint

Please provide more ideas for objective metrics, though we would like to avoid complex metrics due to limited computing time. (VDP2 is very nice, for example, but admittedly too complex for this challenge).

Codecs we will certainly include in the study :

- JPEG (here the reference implementation for JPEG XT, which also allows a backwards compatible 12-bit mode)
- JPEG 2000

Additional codecs welcome, see above. Thus, if you want to contribute, or want to suggest additional codecs, please post here. I'll try to coordinate the contributions from the community with the PCS folks and see what can be done.

In case you're not able to come to PCS, we'll try to find a proxy for you. Admittedly, this year's PCS is pretty remote (Cairns, Australia) for everyone not being and Aussie, and we'll see how to help.