Grand Image Compression Challenge at ICIP 2016

http://jpeg.org/items/20151126_icip_challenge.html

by 112233 • 11 years ago

45 20 11 years ago

vanderZwan3• 11 years ago

Eh... isn't the Joint Photographic Experts Group the one that patents everything?

I'm wondering, what incentive do researchers have to share their new compression algorithms with JPEG? Especially if JPEG patents everything as you suggest.

drpgq• 11 years ago

Can they patent it if you disclose it in a paper?

joosters• 11 years ago

If it works similarly to other standards groups, they don't take the patents from inventors. Instead, they manage the patents and collect fees on behalf of the (many) holders, allowing anyone to buy the rights through a single entity. The various MPEGs and other formats like DIVX work in this way.

TheCondor• 11 years ago

Nope, they don't patent or even hold the patents, they are just a standards group.

fulafel• 11 years ago

Manned by representatives of patent holder companies...

pornel2• 11 years ago

I hope evaluation will use more diverse set of images than the subset given. Only a couple of example images have highly saturated color, and all example images will tolerate poorly done chroma subsampling.

On the web there are categories of images, such as logos, screenshots, renders, large icons, photos with captions, that aren't simple enough to compress well with lossless encoders, but become a mess when compressed with codecs tuned for high-res photos without any sharp highly saturated edges.

ygra• 11 years ago

Well, it is JPEG, after all. So their focus is most likely photographic pictures. As evident by sentences as »contributors to this challenge may want consider contribution of their technology as input to the ongoing JPEG activity«.

vardump3• 11 years ago

You don't need to subsample chroma in JPEG. I often disable it (use 4:4:4 sampling) if there is high color contrast detail. Say a tree with red fruit, marketplace, kid's toys, etc.

pornel2• 11 years ago

I know it's not necessary in JPEG, but some codecs (especially video and therefore video-codec-derived still image formats) have only 4:2:0 option. And if a codec chooses to use it, I'd prefer the test suite to require it done well.

I've recently looked at it closely and found that almost every codec does chroma subsampling incorrectly, but the error is visible mostly in computer-generated graphics, and rarely in photos, and probably that's why nobody cared to fix it.

https://github.com/mozilla/mozjpeg/issues/193

udev• 11 years ago

This would be a good occasion to have scrutinize the compression algorithm in BPG image format http://bellard.org/bpg/ .

vardump3• 11 years ago

For most purposes, image compression is a solved problem.

I assume this challenge exists to create a new image compression format. But I don't see any reason why JPEG2000 doesn't happen again - a new format that almost no one adopts. JPEG was simply good enough.

Although 12-bit color depth with JPEG comparable compression levels would sure be nice.

vanderZwan3• 11 years ago

Holy shit do I strongly disagree with this.

Although 12-bit color depth with JPEG comparable compression levels would sure be nice.

Exactly: there are so many features - like higher bit depth, alpha transparency, different colour spaces - missing form JPEG.

IIRC, JPG2000 wasn't adopted because back in the day we only had Internet Explorer, which required a special plugin, it was too slow, and there weren't many export options.

BPG and FLIF are good contenders in the lossy/lossless area though.

http://bellard.org/bpg/

http://flif.info/

vardump3• 11 years ago

I agree new features would be nice. Like alpha transparency like you mentioned. Or how about cool things like depth or normal maps? I just think almost no one cares. We represent 0.1% of the population.

The last new image format to be universally adopted was PNG in 1996. A lot of entrants have tried ever since, but I think it'd be fair to say none of them mattered the slightest.

Grand Image Compression Challenge at ICIP 2016

Related Stories