What if RED was 2k?

[Pietro Impagliazzo]

Coming from the HDV world is really common to hear the bucket analogy, that more buckets in a tighter area would capture less water.

So the more pixels stuffed in the sensor, the worse your light sensitivity is.

The doubt that crosses my mind is: Would RED benefit for being lower res than 4k?

Would be a crazy light sucker machine that could pull nice images from very dim situations?

Or the RED sensor design would not benefit from being lower res than it is?

I think Graeme has made very clear the OLPF thing, that makes res go down, so If RED was 2k that would mean 1.2~1.4k of actual resolution?

Would that be unacceptable?

How about noise floor?

I'm just wondering.

 

[Tom Lowe]

There was a lengthy discussion of this subject in the EPIC forum. Basically, I'm not sure I really understand it all. For example, a lot of astrophotographers still shoot on the Canon Rebel XT 350D, because it crams fewer pixels (8MP - 6.4µm pixel pitch) into an APS-C 1.6 crop sensor, compared to the XTi 400D (10MP - 5.7µm) or the brand new XSi 450D (12MP - 5.2µm), and so on. I guess in theory this means greater light sensitivity for the 350D.

But one dude came onto the thread and basically argued that while the 8MP sensor might have less noise per pixel, once you downsample the XSi's 12MP to 8MP, you will lose a lot of noise, and it might actually be cleaner than the original, native 8MP image.

As far as RED, I think it's exactly as it should be right now in order to deliver dazzling 2K results. Graeme says RED ONE is actually 3.2K, I think, but others have tested and argued as low as 2.4K, if I recall. In either case, you're covered for 2K finishes, and probably have quite a bit left over if you want to do 3K or 4K filmouts on bigger movies.

It would be interesting if others can post more about this subject. I'm not sure if that other thread ever came to a clear conclusion.

 

[Graeme Nattress]

I think the only person who measured 2.4k (Adam Wilt) realized his mistake, measured again and got figures agreeing with me. It's an easy thing to get wrong. We've also had independent confirmation on CML of my figures.

Say you have 4k, then downsample to 2k, you'll get a better image than if you started with a 2k sensor of the same size, even though that sensor would have larger pixels.

Basically, you want as many pixels as you can, each of as low a noise as you can, but it's the noise per area, for want of a better way of describing it, that counts.

Also, remember Impagliazzo, that the OLPF is necessary for all cameras. Some avoid it to make a higher resolution, but allow artifacts through that they should not.

Basically, making RED 4k gives immense flexibility to the RAW image you'd not have had if we had made it 2k.

Graeme

 

[Pietro Impagliazzo]

Thanks for the explanation Graeme, but my main doubt remains...

What about the light sensitivity? Would it increase for having less pixels?

Considerably?

 

[Graeme Nattress]

Nope, you'd not be more light sensitive. From a simplistic point of view, 2k would have a pixel 4 times the area of a 4k sensor (same area). Yes, it will gather more light. But in that same area on the 4k sensor, we have 4 pixels! We could sum those pixels and get the same recorded intensity as the single, larger area pixel. Or put it another way, the 4k properly downsampled to 2k will exhibit the same noise characteristics, but will be sharper, with less aliasing because we can use better optical and electronic low pass filtering.

Graeme

 

[10s]

Fuji had a super sensor a few years back that couple both large and small pixel sensors into a non square configuration. I believe the large pixels were for low light and small for highlight. Then they used some averaging to cover the the full spectrum. I never followed up on the sensor performance, but I thought the idea was good.

 

[Graeme Nattress]

Yes, Fuji's idea was good, and as far as I can tell, worked well, but got left behind in the megapixel game.

Graeme

 

[jbeale]

Nope, you'd not be more light sensitive. From a simplistic point of view, 2k would have a pixel 4 times the area of a 4k sensor (same area). Yes, it will gather more light. But in that same area on the 4k sensor, we have 4 pixels! We could sum those pixels and get the same recorded intensity as the single, larger area pixel. Or put it another way, the 4k properly downsampled to 2k will exhibit the same noise characteristics, but will be sharper, with less aliasing because we can use better optical and electronic low pass filtering.

I believe this is strictly true only if the A/D sampling noise is negligible (which it may be). As I learned it, there is some noise (uncertainty) associated with each A/D sample. So if you have 4 pixels to sample, you get that sampling noise 4 times, and with one larger pixel you get it only once. Assuming white gaussian noise, when you bin those 4 samples together, the sampling noise is reduced only by a factor of 2, so a single larger pixel wins the noise race. Now admittedly, if other noise terms are much larger than the sampling noise, then this detail becomes relatively unimportant, and I assume that's what Graeme is telling us.

 

[Graeme Nattress]

Quantization noise would be the least of my worries.

Graeme

 

[Carl Larsen]

Nope, you'd not be more light sensitive.
Graeme

Graeme,

I always thought there was a direct (ok, square) correlation between the area of the photosite and light sensitivity. This has always been the explanation I've heard for why the HVX is such a poor low light performer when compared to the DVX... that the same sensor size with more pixels jammed into it equals less light per pixel and therefore lower light sensitivity.

I'm not questioning the validity of your answer, but I am curious on where I'm missing it.

 

[Carl Larsen]

Yes, it will gather more light. But in that same area on the 4k sensor, we have 4 pixels! We could sum those pixels and get the same recorded intensity as the single, larger area pixel. Graeme

Nevermind... I re-re-read what you said and now I get it. Add the pixel values together... I was thinking of what would happen when they are averaged.

As a general concept, however, smaller pixels are still less light sensitive, correct?

 

[Graeme Nattress]

Carl, I was comparing what you'd get from a single large pixel that is the same size as 4 smaller pixels. If you sum the output of the smaller pixels, you'll get the same (give or take, this is a thought experiment) number count as the larger pixel. So, yes, larger pixels collect more light and are therefore less noisy, but not more sensitive or less noisy per unit area, generally speaking.

The problem with the HVX is something I just don't understand as it has larger pixels in it's CCDs than any other "HD" camera as it's only 960x540, which is only very slightly more than you'd get in the DVX, so it's noise levels should be lower than the HD competition at that sensor size.

Graeme

 

[Joshua Provost]

Isn't "fill factor" also a concern? The wasted space in-between pixels? Four small pixels do not necessarily equate to one larger pixel, even when averaged, due to the lost space between those smaller pixels?

 

[Carl Larsen]

Thanks for the reply. You guys are quick...and I thought you were busy working on build 16 or something crazy like that. he. he. Thank you.

Anyway, Per unit area...that makes sense from the thought experiment perspective. But help me understand this: when a 4K image is scaled to 2K, aren't the 4 adjacent pixels brightness values averaged in the scaling process instead of being added together? Otherwise scaling would cause images to become brighter, right?

What I'm hearing you say is that there's a way to get brighter/less noisy images by downressing in post. It that accurate?

Thanks ahead of time.

 

[Graeme Nattress]

Yes Joshua, you would need to take into account fill factor and a whole host of other variables. But that makes things awfully tricky to speak about something "in general".

Graeme

 

[Kevin Halverson]

Isn't "fill factor" also a concern? The wasted space in-between pixels? Four small pixels do not necessarily equate to one larger pixel, even when averaged, due to the lost space between those smaller pixels?

The upper limit of the achievable fill factor is dominated by the topology for most sensors. Specifically, the transistor count and the process used for its fabrication have a considerable effect on the relative area available for the photoreceptor.

Kevin Halverson

 

[Daniel Browning]

I'm that "one dude" who thinks more pixels yield less noise, as in this thread.

Coming from the HDV world is really common to hear the bucket analogy, that more buckets in a tighter area would capture less water. So the more pixels stuffed in the sensor, the worse your light sensitivity is.

The HDV world doesn't have RAW output options, so we have no way of knowing the cameras' true noise characteristics. They also lack S35-sized sensors, bayer CFA, etc., so I think it makes more sense to look at DSLR and other bayer cameras for this discussion.

The doubt that crosses my mind is: Would RED benefit for being lower res than 4k? Would be a crazy light sucker machine that could pull nice images from very dim situations?

No, I don't think it would benefit from it at all. Sensors are all individual things, as Joshua and Graeme said, but you can look at all the products on the market now and see some pretty obvious trends. One of those trends is that sensors with higher pixel density have higher quantum efficiency.

For example, the Canon G9 packs 12 megapixels in 37 square mm. The 5D spreads 12 MP out over a luxurious 864 square mm. That means the 5D has the opportunity to capture 23 times more light (a function of area), assuming a lens with the same t-stop. But it doesn't. And it does not have 23 times less noise. It is far less efficient (per area) than the digicam.

Imagine someone took the Canon 5D sensor and wanted to make a 1/1.7" digicam out of it. So they got an exacto knife out and cut the center 37 square mm out of the 5D sensor and put it in a little camera. On that area, the 5D had 0.5 megapixels. But they're *good* pixels. If you look at the 0.5 MP image at 100%, it's clean and nice, whereas the 12 MP G9 image at 100% looks very noisey. Now resize the 12 MP G9 down to 0.5 MP... the 5D crop looks terrible in comparison. If you don't have an exacto knife, you can do the same experiment by cropping the center 0.5 MP in photoshop out of the Canon 5D. Thus proving that sensors can have *more* pixels in the same area to capture more light and less noise.

If they made the 5D with the same pixels as the digicam, it would have 280 megapixels. And by the time you resize that back down to 12 MP to compare with the real-life 5D, it would have a lot less noise.

The same fact is true for less dramatic comparisons. The 1dsm3 has almost twice as many pixels in the same area as the 5D, but for output sizes, the 1dms3 has much less noise.

In fact, for any camera product line, with few exceptions, the version with the highest number of pixels also has the lowest amount of noise. Like the 350D vs. 450D comparison that Tom mentioned.

People think smaller photosites are worse because they make flawed comparisons, such as viewing 100% pixel size instead of correctly resampling to the same resolution with a sinc algorithm (hint: nearest neighbor is evil), comparing different sensor sizes, or using different processing (two different raw conversion methods). For example, chroma desaturation in high ISO D3 make it appear to have lower read noise than it actually does.

Plus you can compensate for the OLPF in resampling.

...but it's the noise per area, for want of a better way of describing it, that counts. [...]

From a simplistic point of view, 2k would have a pixel 4 times the area of a 4k sensor (same area). Yes, it will gather more light. But in that same area on the 4k sensor, we have 4 pixels! We could sum those pixels and get the same recorded intensity as the single, larger area pixel. Or put it another way, the 4k properly downsampled to 2k will exhibit the same noise characteristics, but will be sharper, with less aliasing because we can use better optical and electronic low pass filtering.

Graeme

Seconded.

What I'm hearing you say is that there's a way to get brighter/less noisy images by downressing in post. It that accurate?

Don't take our word for it; try it yourself. Look at Jim's Build 16... taste images. ISO 1000 downsampled to 0.5K looks noise-free. Look at the 100% crops (4K) and you can see a little noise in them.