V-Nova introduces new codec 'Perseus' that claims is 2-3x better than h265

[hajj_3]

http://www.ispreview.co.uk/index.php/2015/04/new-compression-method-brings-4k-video-streams-to-slower-broadband.html

http://www.v-nova.com/en/products-and-technology.html

I am rather skeptical of this to say the least. Some of the companies they have been working with include Sky TV, Intel and Broadcom.

[roo1234]

Saw this too... Hm, April's fool news maybe? Too good to be true :)
I read it from BBC news...

[nevcairiel]

Doesn't seem too crazy to really be a good april fools. Companies have claimed this before, and then just failed to deliver.

[iSunrise]

Searches on the Web reveal talk about V-Nova an Perseus well before today, so this is certainly not an April fools.

However, I am not sure if releasing this now is such a good idea, when everyone can easily be fooled into thinking this indeed is an April fools. Not sure what they were thinking.

[Nic]

From:
http://www.digitaltveurope.net/346362/v-nova-promise-video-at-audio-bitrates-with-revolutionary-compression-technology/
"The company says its proprietary technology can provide compression or the order of a factor of two or three over existing MPEG technologies, including HEVC, H.264"
"Actmann said that V-Nova’s patents were “clean”. V-Nova plans to license the technology and has no current plans to make it available for standardisation."
"V-Nova claims to use standard off-the-shelf hardware"
"The fundamentals have not really changed. Where the incremental benefits are 10-20%, the business case [for change] is hard to justify. When you show a five times improvement, this is no longer an incremental change but a paradigm shift."

Smells like....well you know what it smells like ;) The scary thing is that there are companies signing up to use this non-standard proprietary unproven codec for their content.....eeeek.

-Nic

[burfadel]

V-Nova's PERSEUS® is an advanced signal processing technology based on the principles underlying human vision.

That makes sense, and why this is possibly 'real'. Basically what it does, going by that quoted line, is cut out all information that you don't perceive when watching the video. So, if you watch a video on a HDTV using this, and then watch a video encoded in HEVC, AVC etc., you won't notice the difference in quality when the Perseus file is say, half the size. If you compared them analytically, side by side, you will probably actually start to see real differences. Basically, it relies on the fact that processing sight is indeed imperfect, and your sight essentially has kind of in-built analogue compression. It's like how when you read, you don't actually read the word! you visualise the first and last letters of the word and your mind 'fills in the gaps'. This was floating around a few years ago. One such 'article' with this information is as follows:
I cnduo't bvleiee taht I culod aulaclty uesdtannrd waht I was rdnaieg. Unisg the icndeblire pweor of the hmuan mnid, aocdcrnig to rseecrah at Cmabrigde Uinervtisy, it dseno't mttaer in waht oderr the lterets in a wrod are, the olny irpoamtnt tihng is taht the frsit and lsat ltteer be in the rhgit pclae. The rset can be a taotl mses and you can sitll raed it whoutit a pboerlm. Tihs is bucseae the huamn mnid deos not raed ervey ltteer by istlef, but the wrod as a wlohe. Aaznmig, huh? Yaeh and I awlyas tghhuot slelinpg was ipmorantt! See if yuor fdreins can raed tihs too.

Even though the Perseus codec isn't about you reading words, it's based around the same principle, from what I believe. They really gave the game away with that quoted line at the top of this post! True, it could be a load of bull, and it might not actually be as efficient or present well as claimed, but in principle there is reasoning behind it. Basically, you're not encoding the source of the image as close to the source as possible, within constraints of the settings chosen, you are encoding purely with the 'destination' in mind at a great expense to the source image. I would imagine the SSIM etc would be bad. You probably wouldn't be able to upscale it too much either, due to the nature of the codec.

[Kurtnoise]

Sounds like marketing bs to me...just read the 1st line from the webpage :

PERSEUS® compresses significantly better than existing codecs (e.g., J2K, h.264/AVC or h.265/HEVC).
too bad for you guys...jpeg2000/h.264/h.265 are not codecs but standards.

:helpful:

[Sulik]

Reminds me of the time there was a announcement about a "magic secret sauce" a company offered that made their MPEG-1 (yes, MPEG-1) have better compression than H.264 :)

[smok3]

I have been secretly developing a new codec, that uses a fixed database size of most common images, so called neverland® technology and calculates deltas from there using C frames (This are b-splined interpolations based on psycho-visual perception®), it beats h.266 by about 120% and Perseus by about 121334378573849%, currently there are secret tests going on with <deleted> on <deleted> by <deleted>.

[Gravitator]

Place Perseus in the collective comparison of quality > http://xooyoozoo.github.io/yolo-octo-bugfixes/#addpsy
Also expect Daala, NGIV, VP10 :)

[Tommy Carrot]

Even if we assume it's not an april fools joke, this still reminds me of the wild claims ON2 made about their codecs. When we were able to try their codecs, we found out that really nothing is true about their claims. Until we can test this codec and verify that there is at least a tiny bit of truth in the stuff said on their home page, i don't expect anything from this codec.

[vivan]

That makes sense, and why this is possibly 'real'. Basically what it does, going by that quoted line, is cut out all information that you don't perceive when watching the video.This is what every lossy codec does, because this is the basics of lossy compression. Every encoder have to optimise for something and the closer that someting to the human visual system is, the better the encoder is. Most are not that good at it (like how they like to blur everything), but some are (x264 and it's psy).

So that line basically tells nothing interesting about their codec.

[ChiDragon]

It's like how when you read, you don't actually read the word! you visualise the first and last letters of the word and your mind 'fills in the gaps'. This was floating around a few years ago. One such 'article' with this information is as follows: [snip]
Not really true; there are plenty of words that share all the same letters and have the same first and last. If that was how we always read, we'd never tell them apart... The type of scrambling also matters for readability.

http://en.wikipedia.org/wiki/Typoglycemia
http://knowyourmeme.com/memes/aoccdrnig-to-rscheearch

[burfadel]

Not really true; there are plenty of words that share all the same letters and have the same first and last. If that was how we always read, we'd never tell them apart... The type of scrambling also matters for readability.

http://en.wikipedia.org/wiki/Typoglycemia
http://knowyourmeme.com/memes/aoccdrnig-to-rscheearch

You can tell them apart because they're not a bunch of random words put together, it also depends on the structure of the sentence. A fluff piece news article isn't going to cover the whole science behind it, if you want to do that you need to read the paper from Cambridge University. That certainly isn't going to be two or three short paragraphs!

If you could read English words, such that you could read out those words aloud, but had no inkling of what they meant, you probably wouldn't be able read the paragraph too well, I'm sure most people reading that paragraph could read it at normal speed and know what it said, if the words were in a different order it would have been very confusing.

[ChiDragon]

you need to read the paper from Cambridge University.
Which doesn't exist.

The meme text was specifically constructed for readability. The Matt Davis page linked at the bottom of the Wikipedia article provides some counter-examples as well as his thoughts on the relevant science.

[foxyshadis]

You can tell them apart because they're not a bunch of random words put together, it also depends on the structure of the sentence. A fluff piece news article isn't going to cover the whole science behind it, if you want to do that you need to read the paper from Cambridge University. That certainly isn't going to be two or three short paragraphs!

If you could read English words, such that you could read out those words aloud, but had no inkling of what they meant, you probably wouldn't be able read the paragraph too well, I'm sure most people reading that paragraph could read it at normal speed and know what it said, if the words were in a different order it would have been very confusing.

That's really more about error correction capability of the brain, anyway. The same facility works with badly damaged video, where we can easily and intuitively piece together what was there, mostly, but there's a reason no one proposes randomly chopping bits out of a stream to save space: It's annoying and ugly. Lowering the resolution and using the maximum quantizer is a less extreme example of the same principle.

Current standards have squeezed most of the theoretical blood out of video; they're chasing smaller and more difficult gains each cycle now. It seems like the only way future standards will ever wildly improve is by getting serious about grain synthesis (which MPEG is loathe to do because it's hard to measure) and long term references (which eats memory and processor cycles for breakfast).