Welcome to our community

Be a part of something great, join today!

Register Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

RED party reel powered by RED RAY

musings on image noise

musings on image noise

Erik Ponti said:
...Banding is not [directly] caused by quantization - banding is caused by the low color depth of 12-bit video... it's impossible to store a gradient, so dithering is required to cover that up. Quantization will remove that noise if there isn't enough bitrate. ...
Click to expand...

Coming from 8-bit video recorders and displays, 12-bit doesn't seem so "low" to me :-). Anyway, do we even know the output depth of Red-ray, didn't someone say 10-bit going to the projector?

At any rate, nothing prevents your codec from representing a gradient internally as a gradient, and maybe wavelet representations effectively do that? The trick is recognizing the presence of a gradient from a (quantized, noisy) pixel raster.

And maybe visually, it's ok to estimate the amount of "random" noise in an image, de-noise it completely, compress that, then on playback, after reconstructing the denoised image, add back the right amount of synthesized random noise.

Now, if the noise was true texture, the output will be wrong and will probably look wrong. But if it was really just grain or noise, and you got the overall amount and frequency spectrum right, you'll never be able to tell the difference on the reconstructed output even in theory (if you don't have access to the true original), because the noise was "random" (a property of the film/sensor, unrelated to the actual scene image) to start with.
 
10bit going to the projector, yes.

Actually the "trick" is a little one only: REDRAY codec works for sure, but the "how it works" speculation is not something I'm going to talk about, other than my usual stock answer of magic pixies. This time we worked them hard, and they seem to have come up with the goods again.

Graeme
 
jbeale said:
At any rate, nothing prevents your codec from representing a gradient internally as a gradient, and maybe wavelet representations effectively do that? The trick is recognizing the presence of a gradient from a (quantized, noisy) pixel raster.

And maybe visually, it's ok to estimate the amount of "random" noise in an image, de-noise it completely, compress that, then on playback, after reconstructing the denoised image, add back the right amount of synthesized random noise.
Click to expand...

Interesting ideas.
 
jbeale said:
Coming from 8-bit video recorders and displays, 12-bit doesn't seem so "low" to me :-). Anyway, do we even know the output depth of Red-ray, didn't someone say 10-bit going to the projector?
Click to expand...
8-bit, 10-bit, or 12-bit, none of them can avoid banding.


Now, if the noise was true texture, the output will be wrong and will probably look wrong. But if it was really just grain or noise, and you got the overall amount and frequency spectrum right, you'll never be able to tell the difference on the reconstructed output even in theory (if you don't have access to the true original), because the noise was "random" (a property of the film/sensor, unrelated to the actual scene image) to start with.
Click to expand...
Quantization already naturally "denoises" this (usually better than specialized denoisers, lol), and you can use debanding filters on playback which do an awesome job... but sometimes they pick up real detail as banding, and dither it... and trust me, that does not look good.

As for storing gradients as gradients... if it was that easy, it would have been done already.

Graeme Nattress said:
10bit going to the projector, yes.

Actually the "trick" is a little one only: REDRAY codec works for sure, but the "how it works" speculation is not something I'm going to talk about, other than my usual stock answer of magic pixies. This time we worked them hard, and they seem to have come up with the goods again.

Graeme
Click to expand...
The lack of visual proof is a bit unsettling...
 
jbeale said:
Coming from 8-bit video recorders and displays, 12-bit doesn't seem so "low" to me :-). Anyway, do we even know the output depth of Red-ray, didn't someone say 10-bit going to the projector?

At any rate, nothing prevents your codec from representing a gradient internally as a gradient, and maybe wavelet representations effectively do that? The trick is recognizing the presence of a gradient from a (quantized, noisy) pixel raster.

And maybe visually, it's ok to estimate the amount of "random" noise in an image, de-noise it completely, compress that, then on playback, after reconstructing the denoised image, add back the right amount of synthesized random noise.

Now, if the noise was true texture, the output will be wrong and will probably look wrong. But if it was really just grain or noise, and you got the overall amount and frequency spectrum right, you'll never be able to tell the difference on the reconstructed output even in theory (if you don't have access to the true original), because the noise was "random" (a property of the film/sensor, unrelated to the actual scene image) to start with.
Click to expand...

I doubt redray adds noise, though it can help images sometimes.

IF redray uses wavelet compression, chances are the wavelet transformation will reduce noise during encoding, since the wavelet "difference" coefficients are likely quantized. This is why some people have said that adding a little grain CAN appear to improve R1 footage, something I have seen myself occationally.

If you want a simple explanation of how wavelet compression works, do a seach for "Haar function". Redcode, jpeg200 etc are more complex than a Haar function, but use the same principles.

You'll see that the wavelet transformation, and the inverse wavelet transformation (used for decoding), aren't what compresses the data, they only convert the pixels into wavelet coefficiets, which are then more easily compressed using existing compression methods ie, simple RLE (runlength encoding), arithmetic coding etc

In other words, the wavelet coefficients are WHAT gets compressed, not HOW.

1) Pixels -> 2) Wavelet coefficients -> 3) compression

I believe the real benefit of converting image data into wavelet coefficients, unlike with DCT systems, is that the samples don't need to be converted from the time/space domain into the frequency domain (computationally expensive, hence is only done in small blocks, hence blocking artifacts..), so is processor friendly, and can therefore process larger amounts of data, hence 4k, whilst still allowing different frequency bands to be processed or analysed seperately.
 
Graeme Nattress said:
REDRAY codec works for sure
Graeme
Click to expand...
That it does Graeme. That it does! I was seriously blown away watching this yesterday. Absolutely amazing! Sitting 2nd row watching the 4k projection I couldn't tell a difference. I asked Deanan if it was Redray codec or Redray codec + Hardware, and basically it was just the Redray codec and playing through their 4k output box [don't remember the name of it off the top of my head]. Also mentioned that some of the details are still being worked out. But as other people have mentioned, Redray worked and it is real. Pretty stunning.
 
Graeme Nattress said:
We'll fill in more details on REDRAY as we go along. What you saw tonight was a technology demonstration to show that 4k image quality is obtainable at practically impossible, or at the very least highly improbable data rates.

With anything visual like this, the proof is in the pudding. I didn't know Ted was going to show the REDRAY demo at that first showing, so when I saw it, I didn't know if it was through the REDRAY codec or not. I could not tell.

Seriously.

It was fantastic to see you all at the event this evening and I look forwards to seeing you all again, and showing you whatever we come up with next!

Graeme
Click to expand...

I just can't tell you how mindboggling that compression is. You're packing 30 times as much visual data as a DVD into the same data rate, when the only other mature (enduser) wavelet codec produces data rates more than 10 times the size.
 
Kenneth Elkington said:
I just can't tell you how mindboggling that compression is. You're packing 30 times as much visual data as a DVD into the same data rate, when the only other mature (enduser) wavelet codec produces data rates more than 10 times the size.
Click to expand...

what is the average size of an NTSC DVD movie file? Like 7GB? I know a lot of 1080p bluray H264 rips now are about 7GBs. that right there shows serious progress.
 
Graeme Nattress said:
10bit going to the projector, yes.

Actually the "trick" is a little one only: REDRAY codec works for sure, but the "how it works" speculation is not something I'm going to talk about, other than my usual stock answer of magic pixies.

Graeme
Click to expand...

Yesterday I was watching a lot of the Reduser event live on Justin's stream. On the side there was a chat going on- and you wouldn't believe how skeptical people there were. It reminded me exactly of when R1 was first announced. It was simply incredible. They kept saying the specs for RedRay were impossible- and this & that about math. blah blah blah is all I heard :)

I also wondered why so many red-doubters/haters would be watching and chatting at a stream for Red?
 
Erik Ponti said:
The lack of visual proof is a bit unsettling...
Click to expand...

Visual proof was provided for those attending. First time round they didn't know they were watching REDRAY. 2nd and 3rd time round, people knew what Ted was up to with his demo and must have been looking at it critically knowing what it was.

Graeme
 
Erik Ponti said:
8-bit, 10-bit, or 12-bit, none of them can avoid banding.



Quantization already naturally "denoises" this (usually better than specialized denoisers, lol), and you can use debanding filters on playback which do an awesome job... but sometimes they pick up real detail as banding, and dither it... and trust me, that does not look good.

As for storing gradients as gradients... if it was that easy, it would have been done already.


The lack of visual proof is a bit unsettling...
Click to expand...

Lack of visual proof? Erik there were 300 or more of us watching this thing on a 4k projector, which by the way, would be the ONLY way to judge it.

I understand YOU did not see it, but I can live with that for now. :)

It was amazing.

Jay
 
"Impossible" is only spoken by those that don't know how to do it...

Jim
 
Jim, I was accused of devil worship, human sacrifice and cabalistic magic last night... All in jest, fortunately. Even if we told people the story behind REDRAY, they'd not believe it anyway :-)

Fact is it worked, looked great and was at an absurd data rate. I love the concept of 4k at absurd data rates!

Graeme
 
Incredible news. When can we expect to learn more about the Red Ray workflow? Or should we try our best to forget it exists while you guys finish up with this whole "camera" and "lens" thing? I admit, it's tough to suppress curiosity when a golden carrot of this magnitude is dangled in front of you.

Eric
 
You must log in or register to reply here.
Back
Top