New codecs

[pieter1976]

CodecSys boosts AV compression
March 08, 2004
By Jason Meserve


A new Codec claims to offer compression rates significantly higher than those used by Windows Media and Real. CodecSys from Broadcast International is currently aimed at business applications such as video conferencing, but has the potential to reduce the storage requirements and delivery bandwidth of multimedia files for live broadcast, Internet streamed video and many other applications.


Codecs such as those from Microsoft (Windows Media) and RealNetworks handle every frame or scene of video with the same compression algorithm. CodecSys analyzes each frame/scene to determine which codec is best suited to compress it. Three consecutive frames of video might use three different compression techniques. Instead of the video being all MPEG-2, it might consist of five different codecs. Audio can be handled in a similar fashion with different codecs handling voice and music.


Broadcast International says its compression is 10 times better than that of MPEG-2, the standard used in DVDs and television. This translates to a 2-hour full-screen movie being stored on a 256MB memory stick rather a 5GB DVD disc, Rod Tiede, president and CEO of Broadcast International, says. He also says CodecSys can deliver high-definition television at 4MB per second instead of the 19.2MB per second used in today's broadcasts.


"Whatever approach you use to deal with digital images, there will always be some trade-off," says Adi Kishore, media and entertainment analyst at The Yankee Group. "By using this approach, they have the best of many worlds."


Broadcast International uses a proprietary software player to decode and display content encoded with CodecSys, but plans to develop it into a plug-in for a mainstream media client such as Windows Media Player so potential users do not have to install and learn yet another application. Tiede also says the company is working on getting the CodecSys decoder embedded into a digital signal processing chip for use in cable boxes, satellite receivers and cell phones where processing power is at a premium.


At next month's National Association of Broadcasters show, Broadcast International is scheduled to demonstrate CodecSys in a $3,000 (around £1650) set-top videoconferencing appliance called Interactive Video. Tiede says the device, which runs on Linux and is H.323 compatible, can deliver full-screen TV video at 100K bit/sec, about one-third the bandwidth that traditional video endpoints use. The 100K bit/sec works only between two Interactive Video endpoints, not any H.323 device.


Despite its promise, CodecSys will not be an overnight success. "Initially it won't replace anything outright," says Michael Hoch, an independent analyst. "What I would like to see happen is compression becoming much more under the covers with something like CodecSys, so that when you go to a customer you're not selling them on MPEG-1, MPEG-2 or Windows Media, you just sell them on your video system, and the customer gets the best compression available."








matrixview (http://www.matrixview.net/)


Adaptive Binary Optimization (ABO©)
ABO© is a new approach to Digital Content Management.

ABO© uses a non-traditional way of transforming data unlike JPEG (Discrete Cosine Transform), JPEG2000 (Discrete Wavelet Transform), JPEGLS, GIF, TIFF, PNG, MPEG1, MPEG2, MPEG4 or MP3.

Such traditional systems depend on the elimination of data to achieve compression, ABO© does not.

ABO© uses the high correlation found in digital content signals and efficiently codes these correlations using our patent pending algorithm. The correlation is efficiently encoded in appropriate bit-planes to achieve significant compression of data. Further compression is achieved by means of standard entropy source coding techniques like Huffman Coding.

ABO© does not require any complex transforms (like FFT, DCT & Wavelet Transforms). It is based on a simple but powerful logical formula. Implementation is carried out with basic mathematical operations.

Unlike existing approaches that generate floating points (decimal values), ABO©'s data is completely binary (integers).




The Results
ABO© will deliver unprecedented quality, clarity, speed and compression at lower complexities and power consumption to
Images
ABO© will deliver images that are smaller in file size, higher in quality and with less errors than existing JPEG and MPEG data.

Sound
ABO© will deliver high fidelity and quality, mathematically lossless sound with higher compression than common formats such as MP3 or any other compression schemes.

Text
ABO© will deliver much higher compression for text based content over existing compression schemes.

[Sirber]

no software yet :(

[pieter1976]

Sirber,

Are you sceptic about the results of matrixview?

[Sirber]

yep

MPEG1/2 at low bitrate = berk. All new codec claim to be the best, none has been so far...

[Marcel]

As long as they don't proof this, they can tell a lot (especially that 5GB / 10 = 256MB...).

Further, the review (http://www.keylabs.com/results/broadcastinternational/40223CodecSys.shtml) linked on CodecSys' page (http://www.codecsys.com) says:

Both clips were encoded at a target bit rate of 1024 Kbits per second. At this rate, we saw no perceptible difference between the native DVD format and the CodecSys-optimized video stream. Resultant compressed file sizes averaged about 1/6th of their original DVD file sizes.
[...]
The 512 Kbit per second encoded video streams were very good as well. Minimal degradation of the video display manifest itself in slight pixilation, that could be easily overlooked[...]
Come on, a sixth of the VOB bitrate without perceptible difference - did they never hear about PSNR & Co. or were they paid not to use them?

I'm not impressed. We had a lot of snowwhites, but none had more than seven dwarfs.

Still waiting for full screen 90° rotated 208*176px*25FPS @100kbit/s playback of content created with freely available tools on Nokia Series 60 devices


edit:
The technology is “Evergreen” in the sense that as new codecs are developed, they can be easily integrated into the library system. Client devices are kept current with updates embedded in the streaming of newly produced content.

So the only step forward within this 'codec' is that it is a kind of codec container inside a video container.

[MfA]

The codecsys thing sounds like a pipedream.

Matrixview could be legit, they have some verifieable claims in some of their docs (mainly compression for CCITT test images, on which the compression they say they can achieve is stunning). Although of course applying it in video coding would be easier said than done even if it did work.

[Joe Fenton]

Originally posted by pieter1976

ABO© uses the high correlation found in digital content signals and efficiently codes these correlations using our patent pending algorithm. The correlation is efficiently encoded in appropriate bit-planes to achieve significant compression of data. Further compression is achieved by means of standard entropy source coding techniques like Huffman Coding.

ABO© does not require any complex transforms (like FFT, DCT & Wavelet Transforms). It is based on a simple but powerful logical formula. Implementation is carried out with basic mathematical operations.

Unlike existing approaches that generate floating points (decimal values), ABO©'s data is completely binary (integers).

In reverse order...

Anyone can do any calculations using integers instead of fp numbers. It means nothing.

Finding correlations in data requires convolution, as any engineer will tell you. Convolution IS a complex transform of the same order as FFT, DCT, or wavelets. This statement is just marketing BS.

Encoding a generic data stream based on mathematical correlation cannot compete with breaking a video into a sequence of images and using knowledge of how the images relate to each other over time.

The Results
ABO© will deliver unprecedented quality, clarity, speed and compression at lower complexities and power consumption to images

ABO© will deliver images that are smaller in file size, higher in quality and with less errors than existing JPEG and MPEG data.

Not very likely.

Sound
ABO© will deliver high fidelity and quality, mathematically lossless sound with higher compression than common formats such as MP3 or any other compression schemes.

Laughable claim.

Text
ABO© will deliver much higher compression for text based content over existing compression schemes.

Not likely.

Overall, these claims seem to be setting up an IPO for a company that will in all probability disappear along with the money. I've seen these kinds of things before. Claims of FMV on floppy disks or other such nonsense. Make a slick presentation to fool the gullible and run before the SEC catches you.

With the claims they are making, they will have to provide substantial proof to back them up.

[MfA]

Here (http://home.student.utwente.nl/m.f.al/abo.pdf) is the patent for what I think is ABO.

Actually no, this must be some type of precursor ... their docs have probability distributions for post ABO values (it uses an additional pass of entropy coding) which shows a zero centered distribution. So they have to be using some type of delta coding too, which the patent doesnt mention.

[Joe Fenton]

I looked over MatrixView's "articles". Funny, they are long on claims, but short on evidence. The most curious thing I noticed is the UTTER lack of references. Not a footnote anywhere. Not a single professor or doctor of Computer Science or whatever signing off on what a breakthrough this is. They claim that their compression is in use at a hospital in Singapore, but a check of the webpage for the hospital finds no references to the product or MatrixView anywhere. Their "presentation" claims the Massachusetts General Hospital has a "pilot program", but again, a search of their web site reveals no references what-so-ever to either ABO or MatrixView.

All very curious tidbits of data. There are no claims from anyone other than MatrixView in the news. The only reference to their products are their own press releases. Considering their earliest releases were in October last year, this is a significant period with no other companies even mentioning this miracle of data compression.

Probably the funniest aspect of the whole thing is this: the only attribution I found in their documents was in attributing the Lena picture to Playboy. I have NEVER seen a data compression paper ever make an attribution on the Lena picture. Yes, it is copyrighted by Playboy, but they have never pressed the issue of missing attributions, and I can't tell you how many scientific magazines and publications have published that picture without attribution in the last two decades.

To this Electrical Engineer, their explanations of their method are vague handwaving and unconvincing. They are geared toward confusing and impressing the PHBs we work for. It would be nice if their claims were true, but I haven't seen one person knowledgable in the field that believes the claims.

[Sirber]

:goodpost:

@Modo

Please close this thread. There is no point keeping it open :)

[Joe Fenton]

Originally posted by MfA
Here (http://home.student.utwente.nl/m.f.al/abo.pdf) is the patent for what I think is ABO.

Actually no, this must be some type of precursor ... their docs have probability distributions for post ABO values (it uses an additional pass of entropy coding) which shows a zero centered distribution. So they have to be using some type of delta coding too, which the patent doesnt mention.

One last thing. I looked over this patent. It's just a claim which I imagine will be rejected if the reviewer has any sense.

The lossless mode is just a bad form of run-length encoding. I've seen better algorithms. The lossy mode is a REALLY bad linear-predictive coding followed by RLE. Neither mode as described in the document would bear out the claims in the document.

Sorry, I believe it even less now. Not going to say more until more "proof" becomes available.

[P0l1m0rph1c]

I mean, these are huge claims that those guys are doing there. They have to show really good proofs to support those claims.

The most likely thing that will happen is that we'll never hear from this company again. These claims have been done by lots of people (have you ever take a look at the comp.compression newsgroup?), and they just disapeared, as they had no proofs to suport those claims.

Anyway, looking forward to news about this...if there will be any...

[CruNcher]

lol hehe remembers me of that company (sorry forgot their name) lol that said they made the invention in Object Based Video Compression and even had huffman saying its a revolution LOL i wonder what happened to them. Website was down after some weeks and nobody ever heared of them again :P

[pieter1976]

cruncher you are looking for this one I think.

pulsent (http://www.pcworld.com/news/article/0,aid,90843,00.asp)

Key to Pulsent's technology is an object-based approach to video compression, as opposed to the block-based approach taken by MPEG-2, according to the company. Pulsent's way is to identify and model structural elements in a frame, instead of what it calls modeling of "arbitrary blocks." Pulsent claims its objects can be far more accurately modeled than arbitrary blocks.

[bill_baroud]

there was the I2BP project too :D
They claimed that with less than 2kb of java code they could stream a video for 56K at DVD quality :D

[CruNcher]

jeah pieter thx for refreshing my memories :)
http://web.archive.org/web/20030217150931/http://www.pulsent.com/

[pieter1976]

found another interesting compression scheme on the internet.

http://www.essential-viewing.com/live/creator.htm

[unmei]

i must be misunderstanding something...
(from the 1st post)
"mathematically lossless sound with higher compression than common formats such as MP3 or any other compression schemes"

do they want to make us believe they can deliver lossless audio ANY bitrate. i mean with mp3 you can have a 16bit stereo 48khz encoding at let's say 32kbit (it sounds terrible admitted). They can encode 16bit stereo 48khz to 32kbit/s _lossless_ (???) wth?

[edit]gah, i'm tired, Joe Fenton laughed at it already :)

[pieter1976]

Matrixview has updated its web site. It all looks very interesting.
Now they have a place on the Australian stock marked and Ernst & Young has done some tests.



matrixview (http://www.matrixview.com)

[Joe Fenton]

Originally posted by pieter1976
Matrixview has updated its web site. It all looks very interesting.
Now they have a place on the Australian stock marked and Ernst & Young has done some tests.

matrixview (http://www.matrixview.com)

WHAHAHAHAHAHAHAHAHA!!!

I can't believe they're still trying to get investors. I pretty thoroughly debunked it earlier (see above). There is NO additional info in the latest press releases or prospectus.

That "Independent Test Result" by Ernst & Young is funny - Ernst & Young is listed on MatrixView's home page as their ACCOUNTANT!! How "independent" is that report going to be?!? They aren't even Engineers! They're CPA's! Suuuuuuurrrrrre I believe them. :rolleyes: :D

E&Y admit in the prospectus that they were being paid for their report, but they don't guarantee any of it.

Here's a quote from the prospectus from E&Y:

The procedures performed do not constitute either an audit or a review and, as such, no assurance is expressed. Had additional procedures been performed, other matters may have come to our attention that would have been reported.

In other words,if they're wrong, they take no blame for you believing them. :D

By the way, that report also said that the JPG and JP2 pictures were done in BEST QUALITY mode, or in other words, worst compression mode. There were no tests against other codecs at all, or in any other mode than Best Quality.

One thing I noticed in the financial section - MatrixView is losing over 4 million A$ a year. They need cash bad. They have no physical assets to speak of. I can't imagine a more risky investment to make.

[pieter1976]

I hope nobody mind that I am keeping this thread alive?

I found a new one on the internet.
They claim to do a better job than mp3

DYNAMAC™ – Algorithmic Compression for Audio, Image, and Video Data

DYNAMAC™ stands for DYNAMics-based Algorithmic Compression. This algorithm, derived from the basics of nonlinear and chaotic dynamics, provides a new compression scheme for audio, image and video capable of delivering a lossless compression ratio of better than 20:1. As the demand for high-quality audio, image, and video over telecommunications links increases, and the available bandwidth diminishes, the solution lies in better compression algorithms. We have just begun to explore the capabilities of this algorithm, yet we’ve developed a tool capable of immediate implementation.

Here is the link.

DYNAMAC (http://www.syncrodyne.com/dynamac.html)

[Joe Fenton]

I don't mind keeping the thread going. It can be rather funny at times. :D

This latest one really doesn't have much info available other than stating it's using non-linear equations to process the data. Non-linear equations aren't solvable as-is, they are usually broken down in some way and processed piecemeal (usually linearly). Depending on the method used, computation can be fairly quick but inaccurate, or take years and be fairly accurate. Major universities run supercomputers for years on non-linear equations to find solutions, so I doubt their PC code is doing the best algorithm. Hard to tell with no real info available.

What I find most troubling about the whole thing was the press release saying they were almost ready back in Oct 2001, then nothing at all until April 2004. Sounds like things didn't pan out the way they planned. Non-linear processing has the potential to do really good things, but until we hear more or see a demo product you really can't say anything one way or the other.