Euclid Video Codec

[3NC0D3_Y0_A$$]

http://www.eucliddiscoveries.com/


EuclidVision is object-based - that is, it segments objects in the video, focuses compression to the most critical and active objects and sub objects, and then enhances the images, resulting in superior video quality. Instead of trying to compress the whole image, EuclidVision identifies objects and sub objects of interest (for example, a person's head, or an eye or ear) in the video while maintaining or enhancing the quality of the background. This approach makes it possible to reduce the size of the file without compromising the video quality.

EuclidVision is a proprietary technology that is also MPEG-4 compliant, and can be integrated seamlessly into existing compatible media players like Microsoft© Windows Media© Player and RealNetworks© RealPlayer, among others.

This technology currently provides a 460% or greater compression improvement over MPEG-4, the standard for digital video, or a 600% improvement over MPEG-2, which is the DVD standard. Euclid engineers are continuing to refine the technology, and EuclidVision has the potential to increase its video compression ratios even more.

Euclid Discoveries' immediate focus is on the creation of a video codec based on the EuclidVision technology. This proprietary codec will revolutionize digital video delivery, enabling users to experience significant improvements in video uploading and downloading speed, and will also provide a number of other significant video processing capabilities.

Euclid is exploring relationships with a number of partners in industries including media content creation and distribution, networking hardware and software, internet portal, search engine and social networking."

Applications

EuclidVision’s current applications are centered around video compression and processing. These applications include the ability to compress and store 100 GB of video in a 10 GB device, HD video streamed to a mobile device in real time, image-based search capability, video enhancement to improve image quality from lower quality video devices, i.e. mobile phones, and the creation of personal avatars from digital images.


Do any of the mpeg-4 codecs implement similar concept alogortihms to the ones mentioned in the euclid video description? (With reference to encoding actual objects on the screen such as a persons head?)

Im sure x264 would benefit greatly from such algorithms.

[buzzqw]

have you ever thought that is most probably total *****?

BHH

[3NC0D3_Y0_A$$]

the website and company look legit.

[unskinnyboy]

Repeat after me - I will use the search function in the future (http://forum.doom9.org/showthread.php?t=109322).

[benwaggoner]

April Fool's Day-1?

Basically, they're describing an advanced form of Differential Quantization I think. We've been doing DQuant in VC-1 for a couple of years, and while it's a nice feature to have around, in the end's it's a perceptual optimization and helps subjective quality some at the expense of a slight drop in PSNR.

No way could you get a 10x improvement with DQuant, even if you were doing perceptually-optimum intra-frame manual tweaks. I could see a really good implementation letting you drop bitrate with typical content by 20% on the outside.

And there's all kinds of content NOT amenible to object tracking. You can be doing a zoom-in on a bunch of M&M's pouring out of the candy machine, and it's still going to need a whole lot of bits to deliver a decent QP, even if you know where all the objects are.

[Dark Shikari]

April Fool's Day-1?

Basically, they're describing an advanced form of Differential Quantization I think. We've been doing DQuant in VC-1 for a couple of years, and while it's a nice feature to have around, in the end's it's a perceptual optimization and helps subjective quality some at the expense of a slight drop in PSNR.

No way could you get a 10x improvement with DQuant, even if you were doing perceptually-optimum intra-frame manual tweaks. I could see a really good implementation letting you drop bitrate with typical content by 20% on the outside.

And there's all kinds of content NOT amenible to object tracking. You can be doing a zoom-in on a bunch of M&M's pouring out of the candy machine, and it's still going to need a whole lot of bits to deliver a decent QP, even if you know where all the objects are.They could be talking about using the "object" system that MPEG-4 ASP offers (which nobody else supports). This allows you to specify the edges of an object, for example, and so as a result all quantization noise outside the object edge is thrown away. I can imagine that, at least theoretically, it might be extremely good for sprite-based video game footage.

[DeathTheSheep]

That thread was made months before this announcement was released: "Euclid Discoveries’ Video Compression Technology Now Works On Movies And TV Shows" --press release. How exciting, a video codec that works on real videos. I'm skeptical... :p

And it has apparently seen development since then. Their website is being updated consistently with propaganda, too. Also "news" about some military technologist from MIT recruited as algorithmic designer. Hmm...

[benwaggoner]

They could be talking about using the "object" system that MPEG-4 ASP offers (which nobody else supports). This allows you to specify the edges of an object, for example, and so as a result all quantization noise outside the object edge is thrown away. I can imagine that, at least theoretically, it might be extremely good for sprite-based video game footage.
Well, the BIFS object behavior is really MPEG-4 systems layer, not tied to ASP in particular. If you had that working for playback, you could use it with VC-1 or H.264 as well.

That said, despite a LOT of effort, no one has ever shown automatic object extraction to really pay off. Object Video claimed to be close back in 2001, and had some interesting demos, but never were able to get even a NDA SDK for independent verification.

Plus, the playback engine gets really painful to do with decent performance, and there's absolutely no ASIC support today. That's a whole lot of blended compositing. And it's easy to imagine scenes where extraction is nearly impossible. Imagine a pan across a football stadium crowd doing the wave. The static-y HBO logo. Any close-up?

Sorenson does have a patent on eye-detection for video conferencing, where they find eyes and use that to figure out where the speaker's face is and encode that less than the background. There's lots of room for techinques like that, but they don't require a new decoder.

I used to get really excited about new codec ideas, but I feel that we're fated to another few decades of finding new wrinkles on DCT + motion vectors...

[CruNcher]

My theory is all the companies in the past involved in such database driven object based video codecs (Pulsent and others) suddenly disappeared and some of them after showing of their stuff when Army guys like from the DOD where in those Demonstrations in the past (i wonder when Euclidvision now disappears) ;)

[Inventive Software]

compress and store 100 GB of video in a 10 GB device

That'll be uncompressed video to a DVD-9 then.

Move along please, I've seen x264 eat those comparisons for breakfast and s**t them before dinner. :D

[DeathTheSheep]

Haha, CruNcher... Like Jan Sloot's mysterious death after his discovery of the 1KB ultimate compression. ;)

"3NC0D3_Y0_A$$" the name is strongly suggestive of conspiracy.

This message brought to you by Enk Ode Yobuttoks.

[CruNcher]

I guess btw Eucliddiscoveries is already on Microsoft's "Who are we gonna acquire next" List @ the very top (tough they should hurry before the DOD got them) :D

[DeathTheSheep]

Haha, has anyone taken a look at their patent they so proudly link to on their main page?

[Atak_Snajpera]

460% or greater compression improvement over MPEG-4,

Yeah, right :) 4 times better than h264. Bull shit .
HQ 1080p @ 3MBps :) hahahahahah

BTW Amigos! Do you remember the whole thing with FairChild codec? The same crap :)