What is 4 times better?

[Dave Sanders]

AllMankind wrote:

Dave Sanders wrote:

ultimitsu wrote:

I would like to ask Mark and everyone else, What do you consider to be "4 x better"?

...compared to the fizzy yellow liquid that passes as mainstream beer. Easily 4X better. And not usually 4X as expensive, so consider it a bargain and support your local breweries!

Most of those good Brewhaus beers are more like 10 times better than that crap the big beer companies like to pass of as beer.

I agree...I was going for subtle understatement.

3 litres of weizen.

My word that's a beautiful sight!

--
Dave Sanders

 

[olliess]

tt321 wrote:

The signal to noise ratio does not scale linearly with the increase of area and collection of light.

You're right, what I said was incorrect. The actual SNR scales as the square root of photons collected, so it is twice as large for 4x the light collection area. What you do get is the noise performance of 4x longer exposure (2 stops more), or more to the point, the noise floor should be 2 stops lower in the scene.

... any advantage in practice cannot scale proportionally. No reasonable maths could get you a "4x better" result in almost any single parameter of the final image.

And yet the measured noise performance between modern generation FF and m4/3 generation sensors does bear out the 2 stop advantage, for the identical illumination at the sensor, for the obvious reasons (it's bigger). That's a completely different argument from deciding what is already good enough performance for reasonable purposes (which I completely agree that m4/3 fulfills).

 

[tt321]

forpetessake wrote:

ultimitsu wrote:

Fellow forum poster Mike Fewster and I were enaged ina discussion of merits of FF over M43. Mike Fewster made an interesting statement:

"What I took issue with was the implication that because the area of the sensor is 4X larger, the picture quality is 4x better. It's better, but not by an amount that can be quantified in direct proportion to the areas of the sensors."

I would like to ask Mark and everyone else, What do you consider to be "4 x better"?

... there is no replacement for displacement.

The 4x bigger area means 4x more light collected, so the same sensor has 4x higher ISO at the same SNR compared to its crop. And it will have 4x more pixels of the same size, thus 2x higher resolution.

Which larger sensor with a 2x diagonal actually has 4x the pixel count as which smaller sensor? This question is across all larger and smaller sensors in the currently available cameras. The answer is of course this set is empty, with zero cameras scaling the pixel count like this. When sensor sizes increase, pixel count does not increase at the same rate, such is the commercial practice of today.

Even if pixel counts scale (i.e. someone were to come up with a 64MP FF sensor), lens engineering will not be able to catch up. So resolution definitely does not and cannot scale.

 

[tt321]

olliess wrote:

tt321 wrote:

The signal to noise ratio does not scale linearly with the increase of area and collection of light.

You're right, what I said was incorrect. The actual SNR scales as the square root of photons collected, so it is twice as large for 4x the light collection area. What you do get is the noise performance of 4x longer exposure (2 stops more), or more to the point, the noise floor should be 2 stops lower in the scene.

Are you talking about one type of noise and ignoring other types?

 

[danijel973]

I would say, from dpreview's quantified data, that medium format digital (for instance that Pentax monster) is four times better than m43. It has significantly better everything you can measure, except size. 35mm, however, is better than m43, but not so much that I would lose sleep over it. It's the same as in a situation I had on vacation where a friend and I had several cameras on the table and I saw a big butterfly on a nearby lavender bush and I simply grabbed the camera with the longest lens, which happeded to be her APS-C Nikon and not my 35mm Canon. Did I lose sleep over it, no. Did I lose the shot? No. There it is:

 

[teohyc]

it's 4 times more likely I'll be bringing out the m43 camera because it's significantly lighter. Also, my audience don't care.

 

[olliess]

tt321 wrote:

olliess wrote:

What you do get is the noise performance of 4x longer exposure (2 stops more), or more to the point, the noise floor should be 2 stops lower in the scene.

Are you talking about one type of noise and ignoring other types?

I'm talking about photon noise. What kind of noise is limiting your sensor?

 

[bluelemmy]

I would like to ask Mark and everyone else, What do you consider to be "4 x better"?

This is not a sensible question. What does better mean? It's like asking 'what is 4 times X' in algebra. Well, that depends on what X is , doesn't it?

If I define better as 'greater surface area', then, yes FF is 4X better.

In order to discuss camera performance you have to have an agreed yardstick. Without that, it's just hot air. Define what you mean by better and ask the question again.

--

David
www.dthorpe.net

 

[boxerman]

Hen3ry wrote:

This stuff is voodoo science.

There's a lot of voodoo, prejudice, non-science, sweeping generalizations and joking going on here. I can provide a decent gloss on a good part of the science.

PROVIDED ONE USES THE SAME SENSOR TECHNOLOGY, you should get around 4 times the number of photons as an upper limit in each cell of the sensor, given the same number of cells (resolution) on a base that is 2X bigger. Saturation number of photons is just proportional to the volume of the cell; the 4 comes from: same depth (from "same technology" assumption), one factor of two for each of "height" and "width" of the cell).

The "shot noise" (random variation just from the fact that sampling a constant but random stream of photons) goes as the square root of the number of photons in the sample. (Easy stats, but I won't explain.) So, a 4x increase in sample size that goes along with the larger size of a full-size sensor results in 1 / (square root of that factor), or 1/2 the noise.

So, POTENTIALLY (see below) you get a decrease of noise, technically defined, by a factor of two. That is not very much by real-world measures. It is the same as reducing your exposure by 2 stops (reducing the number of photons per cell by a factor of four).

But, if you are in low-light conditions already, this factor of 2 can be a killer. For low light, you know, people often would kill for a lens with a two-stop advantage. BUT, if noise is not the issue, e.g., you are in (let's call it) good light, then there just is no advantage AT ALL for the larger sensor, by this noise standard

More caveats: (1) A greater number of photons per cell POTENTIALLY allows you more color resolution (differentiating more shades of color). BUT, color resolution, as far as I know, is limited these days more by technology (are your pix saved with 12 bit resolution or 14?) than by POTENTIAL number of photons per cell.

(2) Shot noise (above) is really simple mathematically, which accounts for the reason people pick it out as "science." But, then, as I pointed out, if noise is NOT the issue, then shot noise is completely irrelevant to (that measure) of quality. Of course, if you LIKE a little noise (certainly has its virtues, in some situations), then noiselessness is a BAD property.

(3) Even for noise, there are other kinds of noise, such read-out noise, which comes from the chip technology and is not so easy to understand mathematically as shot noise is. Read noise can dominate shot noise, so the above simple calculations are also not relevant when read noise is at issue. The 4x better per doubling the sensor, as far as I know, simply does not apply to read noise. I can't tell you the situations under which read noise dominates shot noise. (I could look it up, but so can you.)

(3) Just to re-iterate, this all is under the assumption of constant sensor technology. But, clearly, sensor technology IS changing. So a full-frame sensor from a few years ago just will NOT get that 4-fold increase in saturation number of photons (2x in conventional noise units). I'm kind of thinking that, these days, 3 years, or maybe a few more, roughly is getting you the equivalent of the inherent advantage of FF sensor over 4/3, EVEN IN THE LIMITED DOMAIN OF WHERE SHOT NOISE IS THE DOMINANT NOISE, AND YOU ACTUALLY CARE ABOUT NOISE (which you don't--or shouldn't--with "good light"). So, that 5-year old FF camera will JUST compete with a modern 4/3, by my estimate.

So, there is some real science here, and there is a factor of 4 there...somewhere. But it is just not relevant unless noise is your only measure, AND when noise is actually visible (which it is NOT in "good light").

 

[forpetessake]

tt321 wrote:

forpetessake wrote:

ultimitsu wrote:

Fellow forum poster Mike Fewster and I were enaged ina discussion of merits of FF over M43. Mike Fewster made an interesting statement:

"What I took issue with was the implication that because the area of the sensor is 4X larger, the picture quality is 4x better. It's better, but not by an amount that can be quantified in direct proportion to the areas of the sensors."

I would like to ask Mark and everyone else, What do you consider to be "4 x better"?

... there is no replacement for displacement.

The 4x bigger area means 4x more light collected, so the same sensor has 4x higher ISO at the same SNR compared to its crop. And it will have 4x more pixels of the same size, thus 2x higher resolution.

Which larger sensor with a 2x diagonal actually has 4x the pixel count as which smaller sensor? This question is across all larger and smaller sensors in the currently available cameras. The answer is of course this set is empty, with zero cameras scaling the pixel count like this. When sensor sizes increase, pixel count does not increase at the same rate, such is the commercial practice of today.

Even if pixel counts scale (i.e. someone were to come up with a 64MP FF sensor), lens engineering will not be able to catch up. So resolution definitely does not and cannot scale.

You need to learn to read before answering. I provided pretty obvious explanation of what happens when you crop the SAME sensor. It doesn't have anything to do with the cameras manufactured yesterday or today, the math provides upper bounds of what's possible to achieve. The actual cameras may have many different goals, and constraints: they have different designs, built on different technology, do different processing, etc. And, of course, nobody argued that lenses have to be considered when calculating system performance, and their resolution depend on many other factors and they don't scale as well as sensors. If you are not interested in theory, but only in what is available today, look at the measurements, say DXOmark, the best m4/3 lenses/cameras can achieve about 10MP resolution, while the best FF lenses/systems can be around 20MP. Just don't make your experience the basis for math, it's always the opposite way.

 

[tt321]

forpetessake wrote:

You need to learn to read before answering. I provided pretty obvious explanation of what happens when you crop the SAME sensor. It doesn't have anything to do with the cameras manufactured yesterday or today, the math provides upper bounds of what's possible to achieve. The actual cameras may have many different goals, and constraints: they have different designs, built on different technology, do different processing, etc. And, of course, nobody argued that lenses have to be considered when calculating system performance, and their resolution depend on many other factors and they don't scale as well as sensors. If you are not interested in theory, but only in what is available today, look at the measurements, say DXOmark, the best m4/3 lenses/cameras can achieve about 10MP resolution, while the best FF lenses/systems can be around 20MP. Just don't make your experience the basis for math, it's always the opposite way.

Theoretical upper bounds, when cannot be translated to available commercial products, remain of academic interest only, which is interesting in itself of course.

When cropping a particular shot, then the simple maths works fine, except of course if the crop is located in the corner then your crop is a bit worse than the average for the original and if it's in the centre it's a bit better.

If we accept the DXOmark numbers, then these numbers work out to give FF an advantage over M43 of about 1.4x resolution/sharpness (nowhere near the 4x in the title of this thread or even the 2x predicted by the theoretical upper bounds) and more interestingly the same advantage over APS-C. This is to say that there is zero advantage for APS-C over M43 - The best APS-C seems to give roughly 10 as well!

So if M43 is 100% efficient FF and APS-C would be 70% efficient. Or APS-C and FF have 70% the capability of M43.

 

[tt321]

olliess wrote:

tt321 wrote:

olliess wrote:

What you do get is the noise performance of 4x longer exposure (2 stops more), or more to the point, the noise floor should be 2 stops lower in the scene.

Are you talking about one type of noise and ignoring other types?

I'm talking about photon noise. What kind of noise is limiting your sensor?

In read noise, the equations are different. Larger chips lead to more electronic noise after the photons have done their work. Thus for total noise in the final image the maths is no longer purely 4x and has become approximate. My point is nothing, apart from the quantity of light collected, is precisely 4x and available technology in almost every department has a less than 100% efficient use of that 4x light. By the final image in display, which surely is what matters, in almost every single department the advantage is no longer 4x. This because the larger image circle leads to more compromises in almost everything, optical, mechanical, and electronic and the photon count advantage is eaten away at every stage. It survives in the final image, by and large, but not survives by a rate of 4x for sure.

 

[tko]

FF can be up to four times better in some areas.

But the thought for the day is, given enough light, it's really hard to tell how large the sensor is. There may be some subtleties, but nothing gross.

That where the now classic point and shoot versus medium format comparison came from:

Kidding

Kidding - Luminous Landscape

www.luminous-landscape.com

Tests like this continue to this day. With enough light, benign shooting conditions, and all the time in the world, image comparisons are shockingly close between formats.

ultimitsu wrote:

Fellow forum poster Mike Fewster and I were enaged ina discussion of merits of FF over M43. Mike Fewster made an interesting statement:

"What I took issue with was the implication that because the area of the sensor is 4X larger, the picture quality is 4x better. It's better, but not by an amount that can be quantified in direct proportion to the areas of the sensors."

I would like to ask Mark and everyone else, What do you consider to be "4 x better"?

 

[jim stirling]

tko wrote:

FF can be up to four times better in some areas.

But the thought for the day is, given enough light, it's really hard to tell how large the sensor is. There may be some subtleties, but nothing gross.

That where the now classic point and shoot versus medium format comparison came from:

http://www.luminous-landscape.com/reviews/kidding.shtml

Tests like this continue to this day. With enough light, benign shooting conditions, and all the time in the world, image comparisons are shockingly close between formats.

This rather depends on what cameras and output one is aiming for if the goal is tiny web images then you are spot on. However even with both FF and mFT { best of breeds} at their respective base ISOs and even with benign lighting { God bless it } there is a very noticeable advantage in both

Detail








And in shadow noise











Jim

 

[John King]

Something a bit strange about both these images, IMO.

Completely leaving aside that base ISO for the D800 is ISO 50 (IIRC) and the base ISO for the OM-D is ISO 200 - in the first image there appears to be huge NR or sharpening applied to the OM-D image; in the second, the point of focus of the OM-D image appears to be the closest edge of the object in the lower right of the image, whereas with the D800 image the focus is on the binding of the book.

Can you shed any light on these observations?

Perhaps you could post the original images from which these crops were taken, along with the EXIF data for each of them?

TIA.

--
Regards, john from Melbourne, Australia.
(see profile for current gear)
Please do not embed images from my web site without prior permission
I consider this to be a breach of my copyright.
-- -- --
.
The Camera doth not make the Man (nor Woman) ...
Perhaps being kind to cats, dogs & children does ...
.
I am a Photography Aficionado ... and ...
"I don't have any problems with John. He is a crotchety old Aussie. He will smack you if you behave like a {deleted}. Goes with the territory." boggis the cat
.
Gallery: http://canopuscomputing.com.au/gallery2/v/main-page/

Bird Control Officers on active service.

 

[24hrexposure]

Most of the discussion about sensor size can be summed up as quality versus convenience. When I see someone drinking coffee from 7-eleven, I don't look down my nose at them over my fresh-ground cup straight from the press pot, because sometimes I'm in a hurry and buy coffee at 7-eleven too. These days I count myself lucky to have time even for that.

And now, for my work of fiction:

---

I will here list the performance advantages of my Phase One MF system over full frame*:

• You can get 2.5x (1.3 stops) less noise by keeping the illumination the same and collecting 2.5x the light
• You can (must) have a lens with an iris 2.5x the surface area and more than 2.5x heavier by keeping the illumination the same and collecting 2.5x the light
• You can (must) have 1.6x shallower DoF by keeping the illumination the same and collecting 2.5x the light
• You can expose at 20% the illumination (or use 1.3 stops higher ISO) but still collect the same light, thereby keeping noise the same
• You can have the same density of photosites and have 2.5x the pixels. (The increase in linear resolution is only 1.3x.)

---

Clearly it's pretty silly to compare camera systems on just one dimension. With sensor size, the logical conclusion is to move to medium format, or build your own 8x10 digital back using a flatbed scanner.


The third bullet is significant: capturing more total light for a given composition and shutter speed requires shallower depth of field in the final image, regardless of sensor format. Sometimes you want DoF as shallow as you can get, but other times the foreground and background both have to be in focus. For shots that require larger DoF, there's no difference in the photon shot noise between FF and m43.
Iris size is one of those unfortunate tradeoffs that is non-negotiable, even as technology advances. Maybe some day we'll have FF cameras the size of SD cards, but the lens's front element must remain the same physical size it is today to capture the same amount of light. The stuff in between might get lighter and shorter and the sensor might get smaller (or larger, if you prefer), so eventually maybe we'll end up with a lens shaped like a dinner plate with a little sensor pasted to the back of it.

*Just kidding. Who wants to lug around a huge medium-format camera when there are iPhones? Just kidding. I use the best camera on the market, which is whichever one you, dear reader, own.

 

[John King]

LOL! Spot on!

;-) :-D :-D

--
Regards, john from Melbourne, Australia.
(see profile for current gear)
Please do not embed images from my web site without prior permission
I consider this to be a breach of my copyright.
-- -- --
.
The Camera doth not make the Man (nor Woman) ...
Perhaps being kind to cats, dogs & children does ...
.
I am a Photography Aficionado ... and ...
"I don't have any problems with John. He is a crotchety old Aussie. He will smack you if you behave like a {deleted}. Goes with the territory." boggis the cat
.
Gallery: http://canopuscomputing.com.au/gallery2/v/main-page/

Bird Control Officers on active service.

 

[Great Bustard]

ultimitsu wrote:

Fellow forum poster Mike Fewster and I were enaged ina discussion of merits of FF over M43. Mike Fewster made an interesting statement:

"What I took issue with was the implication that because the area of the sensor is 4X larger, the picture quality is 4x better. It's better, but not by an amount that can be quantified in direct proportion to the areas of the sensors."

I would like to ask Mark and everyone else, What do you consider to be "4 x better"?

Before a meaningful answer can even begin to be given, one would have to define exactly what "4x better" means.

 

[tt321]

Great Bustard wrote:

ultimitsu wrote:

Fellow forum poster Mike Fewster and I were enaged ina discussion of merits of FF over M43. Mike Fewster made an interesting statement:

"What I took issue with was the implication that because the area of the sensor is 4X larger, the picture quality is 4x better. It's better, but not by an amount that can be quantified in direct proportion to the areas of the sensors."

I would like to ask Mark and everyone else, What do you consider to be "4 x better"?

Before a meaningful answer can even begin to be given, one would have to define exactly what "4x better" means.

Mathematically, if something is 4x better, it should be 5x as good. So we are starting on a journey already on the wrong foot...

 

[John King]

tt321 wrote:

Great Bustard wrote:

ultimitsu wrote:

Fellow forum poster Mike Fewster and I were enaged ina discussion of merits of FF over M43. Mike Fewster made an interesting statement:

"What I took issue with was the implication that because the area of the sensor is 4X larger, the picture quality is 4x better. It's better, but not by an amount that can be quantified in direct proportion to the areas of the sensors."

I would like to ask Mark and everyone else, What do you consider to be "4 x better"?

Before a meaningful answer can even begin to be given, one would have to define exactly what "4x better" means.

Mathematically, if something is 4x better, it should be 5x as good. So we are starting on a journey already on the wrong foot...

Spot on :-D