MPEG threatened by object-based video codec

[MxxCon]

http://www.eetimes.com/sys/news/OEG20020322S0105
interesting read..

[dragoman]

Hmmmm.....object-based encoding? Sounds interesting, but this article provides hardly any details (obviously a case where the author has no clue about digital video compression, just repeating what the tech reps told him).

I would like to see more details....

dragoman

Update....I visited the company's website...here's what it says about it's new product:

"We're still in stealth mode. So, there’s not a lot of information we can share at this time. But, stay tuned. We will soon transition from the “best kept secret in Silicon Valley” to the most talked about company in the industry."

I'll let you draw your own conclusions about this one....

dragoman

[Neo Neko]

It would be cool. But it brings back many memories of past charlitans. Time will tell.

[auenf]

MPEG-7 will be very object based, and yes that is next after mpeg-4

Enf...

[MfA]

I think people are thinking too much about MPEG-4 when they hear object based coding ... the way they mention (http://www.pulsent.com/press/Technical%20Overview%203-25-02.pdf) that they can extract objects automatically from natural images I think its more of a region based coding method.

Something like a slightly more advanced version of this (http://vision.ai.uiuc.edu/scyoon/research/proj_video_compression.html).

[Vanos_b]

I heard some time ago, some guys intended to do something like this: detect the objects in the image, use their surfaces as bitmaps and compress them using JPEG/JPEG2000 (this while encoding) and afterwards (while decoding) compose the scene using polygons and apply these images as textures on them (just like in a 3D game). It seems great, but it takes a lot of computer power to encode and it's useless on noisy sources. But the demo (using artificial images, so it was pretty much a game not a video) looked quiet impressive (considering it was a few tens of kilobytes and it worked about 15 minutes).

[ulfschack]

Hope you're not talking about that 60 KB file using directX effects that was circulating a while back. Cause that's all it was, no revolutionary video encoding or rendering techniques were used at all, just DirectX. Admittedly it was a pretty long demo for a file size that small.

BTW I thought mpeg-4 already used object identification.

[MxxCon]

.the .product [fr-08] is the name of that demo
http://www.farb-rausch.com
unless you are talking about something else..

[auenf]

fr08 is all about fitting as much graphics into as little space as possible, all the textures are 'painted' on the fly, rather than stored as compressed images, all that is stored is how to recreate the image.

that sort of thing wouldnt be a benefit to motion video, cause its not made from easily created on the fly teaxtures.

Enf...

[Doom9]

just a few thoughts... how many of you guys have used cce for dvd transcoding? using the manual multipass method you can get away using bitrates that are in svcd like regions and get DVD-like quality (I guess that's what they all broadcast quality).

so, if pulsent can achieve broadcast quality at 1.1mbit/s then this is a way lower improvement than 400%.

[auenf]

i only use CCE for dvd transcoding, and i find that the captures i make (from analogue tv, not from any good quality or special sources) i can use 2000-4000 as the avg bitrate, i leave the max on 9000, but the highest ive ever seen in the files is 5000.

the thing is that pulsent is saying that they can get down to 1.1mbit for broadcast quality object based encoding, i dont think they could get that low without objects in there rather than pixels, we'll have to see tho.

its the same thing with mpeg-7, which will be enhanced mpeg-4 encoding with object based encoding as well, which will be great for web stuff, but objects wont be all that useful for actual motion pictures, unless im reading into this completely wrong (and the article i read a couple of years ago was completely wrong too).

Enf...