[stevebrot]

When we are using only the cropped portion of the sensor your are basically zooming in more to makeup for what was cropped away.

I don't see how that works.

1:1 is always 1:1 and pixel dimensions are always pixel dimensions. A 24Mpx FF capture from a Sony A7III is viewed at the same frame magnification as a 24Mpx APS-C capture from a Pentax KP.1 Direct evaluation of per frame noise should be appropriate with no correction for format. Interestingly, such is also an evaluation per pixel. Since noise is expressed by pixels (they either record noise or signal and not both), is would seem that per pixel evaluation would be more valid. It should be obvious that we are always evaluating per pixel.2

Coming from the perspective of pixel density and per pixel evaluation, the K-1ii at FF has the same pixel density as the K-1ii in crop mode, just more pixels. Per pixel evaluation would show the two as equivalent, but per frame is problematic due to the crop frame having less pixels overall. The absolute amount of noise in the FF frame will always be higher. One could normalize by random sampling of the FF version or simply evaluate per pixel.

The matter of per frame pixel resolution and noise can be problematic. When the K-3 first hit the market many K-5 owners countered its higher resolution by shouting, "the noise, the noise, what about the noise?". Some reviewers responded by downsampling the K-3 images to 16Mpx for comparison. Of course, the noise disappeared with the extra resolution, with the matter of original per pixel noise being swept under the rug.

So...why downrate APS-C sensors based on frame dimensions?

This should also allow you to see that is more that the total amount of light that is captured ( via larger use of the sensor) that determines the amount of noise and DR you can capture.

I would credit the printer driver.


Steve

1 The dot pitch of monitors and printers does not change with change of format.

2 Note that actual evaluation of S/N ratio is not possible since both DxOMark and photons-to-photos employ forensic evaluation of capture data and any notion of signal for a particular frame is an assumption or at best an approximation.

[Ian Stuart Forsyth]

Since noise is expressed by pixels

A pixel as no noise it only has noise if we compare it to other pixels that record the same information from a scene and the deviation from the average
1 pixel compared against 2 will have a different noise variation reading when compared against 1,000
For the K1 in using the FF you have 1 pixel compared to 36mp and the crop 1 pixel compared to 15mp

Since noise is expressed by pixels (they either record noise or signal and not both), is would seem that per pixel evaluation would be more valid. It should be obvious that we are always evaluating per pixel.2

Without relating what that signal represents to the whole does not tell us anything what was recorded
Do you think this is a correct coloration to how the noise is present in the image or that it relates better to how we use cameras in photography ?
Do you think the Nikon D2H has the same IQ as the same exposure values to the K1?
Do you think that the K10 at for the same iso the K10d is only marginally worst thank the K1 iso 400 ?

Or do you think that this is a much better representing with how the images compare.

If you think that it is the first graph that represent what we see in using the camera than many would save a lot of money and not buy the K1

And again here is the K10d better than the K3? If you do iIhave a K10d to sell you

[stevebrot]

A pixel as no noise

I don't believe I said it does. A pixel's value is either noise or not-noise and it records that portion of the frame that it records (at least pre-Bayer interpolation)*. There truly is nothing else but pixels on the output end. Suggestions otherwise (e.g. sub-pixel detail) are the result of wishful thinking. I am not saying you are doing so, but that is the logical extension.

One may plot S/N ratios, but by what means is the signal strength derived short of a best guess or inappropriate interpolation? Remember, this is forensic work and unless one has access to the source (whatever light came through the lens for the capture, the focused image on the sensor) this sort of exercise is pretty much smoke and mirrors. That is where DxOMark may have an advantage in that I believe they are using a projection for their testing under reproducible conditions. They know what went in.

BTW...what are those graphs about and how are readers supposed to know how they apply to the conversation?


Steve

* That is why this sort of testing should always be done on the RAW capture data. The data are not pixels per se but are the stuff from which pixels are directly derived.

[BB_Zone28]

I don't see how that works.

1:1 is always 1:1 and pixel dimensions are always pixel dimensions. A 24Mpx FF capture from a Sony A7III is viewed at the same frame magnification as a 24Mpx APS-C capture from a Pentax KP.1 Direct evaluation of per frame noise should be appropriate with no correction for format. Interestingly, such is also an evaluation per pixel. Since noise is expressed by pixels (they either record noise or signal and not both), is would seem that per pixel evaluation would be more valid. It should be obvious that we are always evaluating per pixel.2

Coming from the perspective of pixel density and per pixel evaluation, the K-1ii at FF has the same pixel density as the K-1ii in crop mode, just more pixels. Per pixel evaluation would show the two as equivalent, but per frame is problematic due to the crop frame having less pixels overall. The absolute amount of noise in the FF frame will always be higher. One could normalize by random sampling of the FF version or simply evaluate per pixel.

The matter of per frame pixel resolution and noise can be problematic. When the K-3 first hit the market many K-5 owners countered its higher resolution by shouting, "the noise, the noise, what about the noise?". Some reviewers responded by downsampling the K-3 images to 16Mpx for comparison. Of course, the noise disappeared with the extra resolution, with the matter of original per pixel noise being swept under the rug.

So...why downrate APS-C sensors based on frame dimensions?

There seems to be many variables to complicate matters when trying to compare and explain noise and SNR between APS-C vs. FF sensors. I'm thinking much of it comes down to the physical pixel size itself when all other things are approximately equal.

Suppose we have two sensors with the same # of pixels which are designed in a similar manner, only one has a smaller sensor size(APS-C) than the other(FF). An optimized design would result in the individual pixel size on the APS-C being smaller to pack in the same # of pixel as those on the FF sensor that would have a larger pixel size to optimize usage of extra area of the larger sensor size.

Using similar on-chip and external IP electronics constructed and shielded in a similar manner could reasonably presume for sake of argument to generate similar noise levels. Having said that, then the individual pixel size on the two different sensors would determine the amount of optical signal(photons) received with the APS-C pixels being less compard to the optical signal(photons) received by the FF pixels by a ratio of the area of their respective pixel sizes when the same exposure it given to each sensor:
(APS-C pixel area) / (FF pixel area)

As a result it's deterministic that the SNR of the APS-C sensor would be reduced at least by the above pixel area ratio. However, you must factor in the derated ISO of the smaller pixel size capturing less light on an APS-C vs. FF that requires a higher gain on the APS-C sensor electronics in order to create the same ISO rating as the FF sensor. This adds additional noise based on the performance of the on-chip electronics.

In this case, the above is a couple of reasons why you would conceivably downrate an APS-C vs. FF sensor when the picture frame dimensions are the same due to the additional noise inherent in the APS-C sensor.
At lower ISO levels the difference may not be very noticeable, but (again, all things being equal) as the ISO is increased the differences become more quickly noticeable. i.e. although not exactly identical, look at the high ISO performance of small-sensor compact digicams vs. larger-sensor DSLRs having the same megapixel count.

---

Last edited by BB_Zone28; 11-21-2020 at 02:22 PM.

[stevebrot]

There seems to be many variables to complicate matters when trying to compare and explain noise and SNR between APS-C vs. FF sensors. I'm thinking much of it comes down to the physical pixel size itself when all other things are approximately equal.

Suppose we have two sensors with the same # of pixels which are designed in a similar manner, only one has a smaller sensor size(APS-C) than the other(FF). An optimized design would result in the individual pixel size on the APS-C being smaller to pack in the same # of pixel as those on the FF sensor that would have a larger pixel size to optimize usage of extra area of the larger sensor size.

Using similar on-chip and external IP electronics constructed and shielded in a similar manner could reasonably presume for sake of argument to generate similar noise levels. Having said that, then the individual pixel size on the two different sensors would determine the amount of optical signal(photons) received with the APS-C pixels being less compard to the optical signal(photons) received by the FF pixels by a ratio of the area of their respective pixel sizes when the same exposure it given to each sensor:
(APS-C pixel area) / (FF pixel area)

As a result it's deterministic that the SNR of the APS-C sensor would be reduced at least by the above pixel area ratio. However, you must factor in the derated ISO of the smaller pixel size capturing less light on an APS-C vs. FF that requires a higher gain on the APS-C sensor electronics in order to create the same ISO rating as the FF sensor. This adds additional noise based on the performance of the on-chip electronics.

In this case, the above is a couple of reasons why you would conceivably downrate an APS-C vs. FF sensor when the picture frame dimensions are the same due to the additional noise inherent in the APS-C sensor.
At lower ISO levels the difference may not be very noticeable, but (again, all things being equal) as the ISO is increased the differences become more quickly noticeable. i.e. although not exactly identical, look at the high ISO performance of small-sensor compact digicams vs. larger-sensor DSLRs having the same megapixel count.

I am glad you understand all that you have written. I was not able to follow the logic of assuming a larger pixel diameter for FF at the same pixel pitch as APS-C based on format alone, so you lost me in the end. Thanks for taking the time, though.


Steve

[BB_Zone28]

I am glad you understand all that you have written. I was not able to follow the logic of assuming a larger pixel diameter for FF at the same pixel pitch as APS-C based on format alone, so you lost me in the end. Thanks for taking the time, though.
Steve

Are you referring to the 'pixel pitch' in the displayed image or the 'pixel pitch' on the surface of the physical sensor?