as before, I continue here to announce (inofficially) news from the ISO WG1 committee, also known as JPEG. For official announcements, please wait for the JPEG press release on www.jpeg.org. Or subscribe to firstname.lastname@example.org via https://listserv.uni-stuttgart.de/ma...info/jpeg-news.
First, on JPEG XT: This is a backwards compatible extension to the old JPEG image compression system, extending it to higher bitdepth, high dynamic range, lossless coding and coding of alpha channels. JPEG XT part 3 defines a (also backwards compatible) boxed-based file format on which the rest of the parts are built. It became "International Standard" this meeting, thus work is done!
Parts 6 and 7 (higher bitdepth, floating point) became "final draft" and will be ready in October. Part 8 (lossless) and part 9 (alpha channels) entered "Draft International Standard", and we can likely progress them to "International Standard" in October as well.
Not much changed, actually. The order of the DCT filtering changed in part 8 to align it with MPEG C (ISO/IEC 23002-2), Parts 6 and part-7 profile C got an additional (optional) downscaling step, and parts 7 and 8 fixed an issue in clipping components.
The reference software entered "committee draft" stage, and an updated version of it will be available on www.jpeg.org soon. Despite integrating the (small) changes in the standard, it also includes some features from MozJPEG to improve performance: A new parameter will allow you to select the quantization matrix, and a deadzone quantizer improves the rate/distortion performance. I also played with the MozJPEG trellis quantizer, but the results were comparably to the deadzone quantizer, though a lot more image dependent at a much higher complexity, so I will currently keep this in a side branch and not in the reference. I will announce here as soon as the software is on www.jpeg.org.
JPEG also started a new initiative on lighweight frame compression, targeted at video links and cable connections. As I posted in the last meeting, we now really confirmed the name of this work item, it will be "JPEG XS" as in "extra speed". It is not backwards compatible, and aims at a 1:6 visual lossless or 1:2 lossless compression with "as few gates as possible". Interested parties may want to subscribe to https://listserv.uni-stuttgart.de/ma...info/jpeg-news. For that, you shortly need to introduce yourself to get an invitation.
JPEG continues to explore future work items, in particular compression of point cloud data and plenoptic data (lightfield camera data), also holographic data. We had a workshop on this, and while this sounds all pretty fascinating stuff, we did not yet start a work item on this and currently just explore the possibilities and market needs. https://listserv.uni-stuttgart.de/ma...eg-innovations is a public mailing list for those of you that are interested.
Other than that, some maintenance work on JPEG XR and JPEG 2000. (Now that we have JPEG XR,XS and XT...)
Last but not least, we also reviewed the image compression challenge at PCS 2015 in Australia. Objective and subjective tests have been made, including BPG, HEVC-IFrame, JPEG 2000 (optimized), JPEG, JPEG XR, VP9 and some I might have forgotten. Interestingly, BPG works a bit better than HEVC-Iframe as it avoids some of the crude tools from HEVC, and is ahead of JPEG 2000 (15 years old by now). JPEG XR and JPEG are not too far apart (no suprise). Interestingly, for some metrics the Mozilla Daala codec works really surprisingly well. This advantage does not quite show in subjective scores, which is a result that still has to be investigated. But at least, congratulations Daala!