[Rondec]

Ok,
lets say we have the 36 MP Sony FF sensor.
We cut out the middle of it to get an APS-C sized one.
Now we shoot the same scene.
Since on the APS-C sensor more light hits a single pixel, it has a higher equivalent base ISO?

I think the whole point is that you have to normalize your output. The viewing/print size should be the same whatever sensor you are using and yes, in that case, cutting the middle out of your 36 megapixel sensor is at a significant disadvantage as compared to taking the whole sensor's output.

In low iso situations or small print sizes, it probably won't be evident, but the more you push it, the clearer it will be.

[derekkite]

A quick reading indicates iso isn't a measure of the sensitivity of a pixel; it's well depth and efficiency. It is with a given lens and aperture the exposure of a scene at a given luminosity what gives you 18% grey or 100% white or saturation. Base iso is the sensor ability to gather light before any amplification is applied.

So to test the iso of the sensor would require a lens and luminant scene, set the body to the lowest iso and see what luminosity is required to get 18% grey and saturation. Same lens on another body, same test. For color, a bit more complicated, essentially plotting the results on a curve.

Luminence used in the iso calculaton is a measure of intensity / sq m.

So a larger sensor at the same scene luminosity and same lens would be absorbing light from a larger area of the scene, and since all the calculations are light/area, the larger area of the sensor would affect the iso number.

So if I'm shooting a certain scene, iso 200 say, apsc and ff and mf, with the same sensor technology other than size, the internal sensor gain applied to produce that iso 200 will be different.

Is this right?

[ogl]

I also have a specialist degree that utilizes a lot of math and physics, but my education doesnt make me more right then others. The math we are talking about here are on the level that 12-14 year old kids learns.
.

I doubt that you had learned film speed measurement systems when you were 12-14 years old

https://en.wikipedia.org/wiki/Film_speed

[blende8]

Is this right?

That is the question.

In my view you must shoot the same scene, so using different distances for different sensors.
Everything else constant: Same lens, same aperture ...

Since you are shooting the same scene, it is always the same amount of light hitting the sensor. But for different sensor sizes, the photon/pixel ratio will be different.
The smaller sensor is getting more photons/pixel, thus having a higher base ISO.

The interesting thing about this is that despite using the same sensor material for the different sensor sizes, still the base ISOs are different.

[Rondec]

I think the problem which I see with all these discussions is that folks look at them on a pixel level and on a pixel level, the D800 and K5 will be exactly the same with regard to noise, but if you go to print a K5 and D800 print, you will definitely see a difference as long as you print them at the same size.

[Kunzite]

I think the whole point is that you have to normalize your output.

Why?

[Simen1]

I doubt that you had learned film speed measurement systems when you were 12-14 years old

https://en.wikipedia.org/wiki/Film_speed

I did not learn about film speed when i was 12-14, but i didnt claim so eather. I wrote thats the level the _math_ are at. Please stop misunderstanding on purpose.

[ogl]

I did not learn about film speed when i was 12-14, but i didnt claim so eather. I wrote thats the level the _math_ are at. Please stop misunderstanding on purpose.

[falconeye]

I just disagree in using equivalent ISO in comparisons. When some sensors have larger well depth then others then it just means they have larger pixels. Max charge per unit area is usually not that different between sensors.

In your hurry to disagree, you probably miss the whole point.

The well depth of the D800/E is 2200 e-/µm^2,
the well depth of the D810 is 3300 e-/µm^2.
[source: Sensorgen.info data for Nikon D810 which in turn is a better look at DxO and its individual measurements -- except for older entries which are typically broken (show false data).]

That is considerably (50%) more and you need a lower iso setting to make proper use of it (to not underexpose). Therefore, comparing both cameras (or D810 and 645Z) at iso 100 is a bad way to compare their capabilities.

My way of doing sensor-size independent analysis did cope with that fully though.

How do you establish/calculate this?
...
And this is bad because ...?

Equivalent iso is defined by many sources, e.g., LumoLabs -- Camera Equivalence -- Whitepaper
Equivalent base iso is established by first measuring a camera's true base iso, e.g., using DxO measurement called "ISO sensitivity" (e.g. Nikon D810 : Measurements - DxOMark ). For the D810, it yields ISO 47.

And then scaling it with crop^2 to arrive at the 35mm-equivalent base iso.

Assuming equal quantum efficiency, equivalent base iso is a direct measure of the sensor's overall (total) well capacity. The nice thing is that this is all one needs to know about the intricate ISO norm which defines camera sensitivity.

However, older sensors have smaller quantum efficiency, a fact which I ignored here for the sake of simplicity. In current age, all CMOS sensors, incl. Canon, are between 50% and 65% Q.E (more typically, between 55-60%). To avoid the effect of Q.E., you may want to use SNR at base ISO instead (D810: 46.3 "DxO"-dB). But IMHO, just using the value of 35mm-equivalent base ISO instead is good enough. These days, Q.E. is improving slowly only.

> "And this is bad because"

This is bad because it means that the maximum SNR the camera can produce is rather limited (37.5 dB for the RX100m4). This means less clean (10dB less) images although physics would allow for better even for a small sensor.

P.S.
Anecdotical comment ... the RX100m3/4 has an in-camera app (badly named "Smooth Reflection") which allows to stack a (large) number of shots in-camera and in RAW (tripod required) which nicely simulates very low 35mm-equivalent base ISO which can become as small as ISO 2.3 (takes between a few seconds and 4 minutes to produce). Loading such a RAW in Lightroom is jaw-dropping

[Rondec]

Why?

Because the assumption has to be that your goal output is the same size, regardless of the size sensor you are using.

[blende8]

I think the problem which I see with all these discussions is that folks look at them on a pixel level and on a pixel level, the D800 and K5 will be exactly the same with regard to noise, but if you go to print a K5 and D800 print, you will definitely see a difference as long as you print them at the same size.

No.
According to my logic the K5 should produce the better result, because every single pixel gets more light.

I am not sure if I am right, it does not sound right, but I do not find the error.

[RonHendriks1966]

In your hurry to disagree, you probably miss the whole point.

The well depth of the D800/E is 2200 e-/µm^2,
the well depth of the D810 is 3300 e-/µm^2.
[source: Sensorgen.info data for Nikon D810 which in turn is a better look at DxO and its individual measurements -- except for older entries which are typically broken (show false data).]

That is considerably (50%) more and you need a lower iso setting to make proper use of it (to not underexpose). Therefore, comparing both cameras (or D810 and 645Z) at iso 100 is a bad way to compare their capabilities.

My way of doing sensor-size independent analysis did cope with that fully though.

on the well depth. In e-/ųm^2

K-7 1025
K-5 2038
K-01 1702
K-3 1612

We all experienced how bad the K-7 was, but what is the deal with K-3 (or is this compensatie with having 50% more pixels?)

[clackers]

We all experienced how bad the K-7 was, but what is the deal with K-3 (or is this compensatie with having 50% more pixels?)

Correct.

The K-3's.pixels generate noisier images (note that this does not having *anything* to do with sensor format, unlike Falconeye's equivalence babble).

But those 24Mp downsample nicely to compensate.

[Kunzite]

Equivalent iso is defined by many sources, e.g., LumoLabs -- Camera Equivalence -- Whitepaper

Self-referencing to support your own claims? Really?

[falconeye]

on the well depth. In e-/ųm^2

K-7 1025
K-5 2038
K-01 1702
K-3 1612

We all experienced how bad the K-7 was, but what is the deal with K-3 (or is this compensatie with having 50% more pixels?)

I checked one of your results (K-3, 1640 e-/µm^2). So, looks correct. Anyway, these numbers are derived from DxO measurements by fitting a physics model. I'd take them with a grain of salt, systematical errors are at least 10%.

And no, it isn't compensated by more pixels as you already normalized per unit area. The K-5 seems to have more well capacity and it shows in the DxO landscape score. BTW, smaller pixels do not, by themselves, lead to smaller or larger well capacity per unit area. Rather, it is a result of pixel design, silicon design rules and the manufacturing process. More well capacity increases cost of manufacture.

---------- Post added 28th Dec 2015 at 20:48 ----------

Self-referencing to support your own claims? Really?

I make no claims, just wanting to help. Moreover, the referenced paper is cited by enough independent sources [e.g. https://luminous-landscape.com/dxomark-camera-sensor/ ]