[colonel00]

If you were paddling your canoe across the desert and the wheels fell off, how many pancakes would it take to shingle a doghouse?

Since its the answer to the ultimate question of life, the universe and everything...I'm gong with 42

[clackers]

AFAIK a larger sensor would have an advantage regarding the shot noise, no matter the technology used.

(Laughs)

Do you think the shot noise equations refer to the whole sensor?

They do not.

They refer to an individual pixel. ☺

Each pixel has an electron well capacity rating ('pixel pitch' is a proxy for this), the sensor itself doesn't, it's just a collection of pixels.

Total sensor size is a crock, but like any good myth, is pervasive.

[normhead]

Since its the answer to the ultimate question of life, the universe and everything...I'm gong with 42

I thought it was 39....

---------- Post added 01-02-16 at 04:12 PM ----------

But doesn't this translate to "bigger sensor... does"? In other words, a bigger sensor can accommodate more pixels of a given size, or, each of a given number of pixels can be larger if the sensor area is larger.

Only if you take a different picture. If you stop down your bigger sensor one stop to create the same picture you are going to create with the small sensor, you have twice the intensity in the small sensor on half the size. To take the same image, total light is the same.

This image was taken with a sense on a bridge camera, that's half the size of an APS_c sensor. It doesn't matter whether you shoot APS-c or 36mil. If you shoot the same DoF, you use the same amount of light to get the same image. And same total light means same signal to noise.

The light intensity in the Lumix is 4x that of the intensity on the FF image. The APS-c image is taken with 2x the light of the FF image. What the FF enables you to do, is to shoot for 4 times less DoF than the Lumix or 2x less DoF than APS-c at it's widest setting.

36mm sensors often have better AF and can have smoother out of focus areas at 50mm and less, but between those two things and the above, that's pretty much it. It hs just as much detail as a K-3 or 24 MP 36mm camera would have, and the same DoF (pretty impressive given that it's a 20MP sensor, but it's newer technology.) . Notice the soft areas in the orange area under the bill of the bird. It could have used a little more DoF. You couldn't have improved this image at all, using APS-c or 36mm of the same MP. Same image, same total light. That's what the theory of equivalence really means.

IN this case and this case only, (there are other cases to be considered when selecting camera equipment), using more total light on a 36mm sensor, would have produced less depth of field and a less pleasing image. Everything affects everything else. it's never as simple as some people make it out to be.

[clackers]

.

The screen graphs shows us that pixel for pixel both sensors perform roughly the same. So an APS-C crop from the 810 will be just as good/bad as a photo from the K5.

The print graph shows us that when the 810 uses the full frame it will outperform the K5. Not because it has bigger pixels (which it doesn't), but because it has a bigger total sensor area.

Then ... you genuinely don't understand downsampling.

The print graph is the result of post-processing, and only pixel numbers count.

A K-3 will downsample *better* than a Canon 1D X.

Comment?

Let me explain what downsampling does.

It's about taking the original picture from the camera, noise and all (you at least see that the pixels start off the same in the FF and APS-C).

When DXOMark do the downsample to 8Mp, one out of two pixels from the K-5 need to go, and two out of three from the 810.

They don't do this randomly.

Three adjacent pixels are looked at, and an algorithm weighs up what to replace them with. Blue, blue, yellow might be replaced with a single blue to represent the group. That's how individual noise pixels are discarded. Overdone, noise reduction wipes out all individual details too, so that human faces look like wax dummies.

But you get the idea. And all the software uses is many pixels, and does not know or even care of the size of the sensor they came from.

[clackers]

You obviously contradict yourself, The first line above is incorrect, but the second line is correct.
What do you think happens with sensor size on the second sensor if you use same pixel design, where the first sensor has 16MP and the second sensor has 125% more pixels? And how does the extra 125% pixels affect image noise?

Baffling, Fogel.

You just said it yourself: "Same pixel design".

It is always about the pixel size for the image capture and pixel numbers for the post processing.

A large sensor does not have to have more pixels.

A Nikon D750 has no more pixels than a K-3, and a Canon 1D X has less.

The Canon 5DS has so many smaller pixels crammed into it that a lot is expected of downsampling to compensate, and no one is impressed with the results.

[Rondec]

Clackers, I would just say somehow you have to compare sensors with different sizes and different resolutions. If you insist on only looking at things on a pixel level, then yes, sensors with higher numbers of pixels will be at an inherent disadvantage. A Sony A7s will wipe up the floor with a D810, for example on a basic pixel level.

But that makes no sense. Because you have to assume some common end print or viewing size. Either you have to up sample the sensors with less resolution or down sample the sensors with more resolution. It doesn't really matter which you do, but that's what the print button on dxo mark does -- down sampling to the same megapixel image.

The whole point of photography is image creation. The question is how does an individual sensor perform at a given size image (pick whatever size you want). Comparing a 4 by 6 image shot with a D700 to a 16 by 24 shot with a D810 is an odd way to try to figure out which sensor will perform better in the field.

[clackers]

If you insist on only looking at things on a pixel level, then yes, sensors with higher numbers of pixels will be at an inherent disadvantage.

No, Vincent, I'm not sure you're following the narrative.

"Quantity has a quality all of its own".

Downsampling permits the K-3 to return noise performance not a lot worse than the K-5.

It is a good thing in of itself.

But, again, it's *not* due to sensor format.

---

Last edited by clackers; 01-02-2016 at 03:00 PM.

[Ian Stuart Forsyth]

No, Vincent, I'm not sure you're following the narrative.

"Quantity has a quality all of its own".

Downsampling permits the K-3 to return noise performance not a lot worse than the K-5.

It is a good thing in of itself.

But it's *not* due to sensor format.

Just to let you know the K5 and K3 are the same format so bring this up?

[clackers]

Just to let you know the K5 and K3 are the same format so bring this up?

Exactly, Ian.

Downsampling is based on numbers of pixels only.

It is a mathematical algorithm, like lossy compression.

[Gimbal]

Then ... you genuinely don't understand downsampling.

The print graph is the result of post-processing, and only pixel numbers count.

A K-3 will downsample *better* than a Canon 1D X.

Comment?

Let me explain what downsampling does.

It's about taking the original picture from the camera, noise and all (you at least see that the pixels start off the same in the FF and APS-C).

When DXOMark do the downsample to 8Mp, one out of two pixels from the K-5 need to go, and two out of three from the 810.

They don't do this randomly.

Three adjacent pixels are looked at, and an algorithm weighs up what to replace them with. Blue, blue, yellow might be replaced with a single blue to represent the group. That's how individual noise pixels are discarded. Overdone, noise reduction wipes out all individual details too, so that human faces look like wax dummies.

But you get the idea. And all the software uses is many pixels, and does not know or even care of the size of the sensor they came from.

Thanks, but I do understand downsampling, and it's only necessary when our displays or printers has a lower resolution then the photo we want to display. But lets say our printer and paper is capable of infinite resolution so we can print a 50Mp picture without down sampling on a A4 paper. And all the individual pixels are there, but we will need a microscope to se them.

So what will happen? Your eyes will downsample the picture for you, simple because they don't have enough resolution to see the individual pixels. The result will be the same as when the computer downsamples, details (and noise) will be lost. Now, is it still cheating?

[clackers]

Thanks, but I do understand downsampling

(Cough).

You would not have said those things if you did.

Now, is it still cheating?

See my reply to Rondec.

1. Downsampling is entirely legitimate.

2. Downsampling is due to number of pixels available to the software, and *not* sensor format. The algorithm will do its job on an image from a 645Z or an iPhone.

If you have proof that the resampling noise reduction slider from Lightroom or any other product pays attention to sensor size, I'd like to see that. :-)

[Ian Stuart Forsyth]

Exactly, Ian.

Downsampling is based on numbers of pixels only.

It is a mathematical algorithm, like lossy compression.

It can be easily shown that the size of the sensor has a direct impact on the noise in an image, If we use your pixel level metrics at DXO you can that the K3 gathers less signal for that portion of the sensor used.


When the screen view is selected what we are in essence doing is comparing a 16mp crop of the K3 sensor to the full 16mp k5 sensor ( viewing at 100 % ) with no scaling and you can see that the noise performance of that K3 crop drops by the factor that we have cropped the k3 sensor, a factor of about 1.224 times.
If we compare the K5 at iso 150 and the K3 at iso 100 they are very close to noise performance demonstrating that the size of the has a direct influence on how much noise would show up in the image ( this is without any scaling).

[Gimbal]

(Cough).

You would not have said those things if you did.



See my reply to Rondec.

1. Downsampling is entirely legitimate.

2. Downsampling is due to number of pixels available to the software, and *not* sensor format. The algorithm will do its job on an image from a 645Z or an iPhone.

If you have proof that the resampling noise reduction slider from Lightroom or any other product pays attention to sensor size, I'd like to see that. :-)

No one has ever said that downsampling has anything to do with sensor size (except you).
So since we all understand downsampling, what is your problem with it?

[TwoUptons]

If you were paddling your canoe across the desert and the wheels fell off, how many pancakes would it take to shingle a doghouse?

None, because ice cream has no bones.

Though I am expecting the new K1 to have better noise characteristics and more resolution than the bright yellow one I shoot with now...

- Eric

[RonHendriks1966]

None, because ice cream has no bones.

Though I am expecting the new K1 to have better noise characteristics and more resolution than the bright yellow one I shoot with now...

- Eric

What.......The K-1 isn't going to be available in yellow?