[Rondec]

This whole discussion gets a little weird. If you have a D800 and a K5 II, you have pixels that are exactly the same size and exactly the same quality. Everyone knows that. The difference is that you have 36 megapixels of them with the D800 and only 16 megapixels of them with a K5 II. The question then becomes what happens if you print a photo at "x" size made from each sensor. Below a certain size, you probably won't see a difference and above a certain size, you most certainly will see a difference. Arguing about it seems silly.

It is understood that at low iso in good light, all of these cameras are awesome, but there is no doubt that in low light and high iso situations, full frame cameras do better. Why this is even an argument is beyond me.

[mecrox]

This whole discussion gets a little weird. If you have a D800 and a K5 II, you have pixels that are exactly the same size and exactly the same quality. Everyone knows that. The difference is that you have 36 megapixels of them with the D800 and only 16 megapixels of them with a K5 II. The question then becomes what happens if you print a photo at "x" size made from each sensor. Below a certain size, you probably won't see a difference and above a certain size, you most certainly will see a difference. Arguing about it seems silly.

It is understood that at low iso in good light, all of these cameras are awesome, but there is no doubt that in low light and high iso situations, full frame cameras do better. Why this is even an argument is beyond me.

Exactly, and especially when most cameras have got so good that 'low light' only begins at ISO 1600 or 3200, sensitivites which were seen as completely amazing only a few years ago. In fact I've had better results in low light from a camera with a smaller sensor because of very good IBIS and a lens which is excellent at f1.8 to f2.5 whereas the larger camera's lens was very mediocre wide open and its stabilization (if any) inferior, so I can put the stops saved into a lower ISO. There is more than one way of doing this stuff.

[bxf]

The same arguments are presented over and over and over. The one thing that I am repeatedly failing to understand is Clackers' insistence that DXO's SCREEN tab is the relevant one. Sure, I understand his argument with respect to pixel/RAW noise, but in the end, one has to perform comparisons on the end result, i.e. a print or screen image. In either medium, the images from the different sensors must be produced to the same given size in order for the comparison to be meaningful. As I understand it, such a comparison is represented by the PRINT tab on the DXO graph.

Why is this even arguable?

PS It is not my intention to disparage Clackers. I just happen to be unable to see the logic of his argument on this matter.

[rawr]

Life will be so much easier when, after a few years, everyone here has a K-1, and some identical lenses in their camera bag (eg the D FA 28-105) ...

[jsherman999]

That... does not make sense. What are you meaning here by exposure?

Same f-stop and shutter speed at the same FOV. Why this happens is because even though it's the same exposure, the physical aperture (the lens iris opening) is larger on the FF combo.

For example:

50mm f/2.8 (FF) == 50 / 2.8 == 17.85mm physical aperture
35mm f/2.8 (aps-c) == 35 / 2.8 == 12.5mm physical aperture

This link talks about it more if you want a good intro: a good read on the subject from dpreview

In this context, I take it as the light volume registered per square mm of sensor - it's independent of total sensor area. But assuming the same light volume per square mm and the same iso setting, then the noise level per pixel will be the same.

Yes, you would be correct - the light density (or intensity if you prefer) at the same f-stop is the same (light volume pr sq/mm,) but the total light being used to create the image is more because of the larger physical aperture needed for that FOV. More light being used to create the image results in less noise in the overall image.

Each square mm of sensor doesn't know if it's part of a large sensor or a little sensor - it generates the same sensor noise regardless. (Although, if they're cut from the same wafer, the larger sensor has -more pixels-. So if you downsize the result to the same dimensions as the smaller sensor, sure, the larger sensor has less noise. You're losing noise with the downsampling (along with resolution). But if you're not downsizing, then the noise in each 100% crop is inherently the same, because it comes from the sensor.)

The greater amount of total light being collected by the larger sensor area is what causes the image to surpass the smaller sensor - if you re-sample more MP down to a normalized image, you can't create light - in other words, if re-sampling worked that way you would be able to create an apsc- sensor with 100MP, and the re-sampling down to 16MP would beat every native 16MP aps-c sensor out there, simply by re-sampling.

What you can and should expect is for sensors of the same size and similar gen, re-sampling the higher-res one down to the lower-resolution matches the lower res one, even if the per-pixel performance of the higher-res one is worse - and we have multiple examples of that, like the K3 vs. K5. When one sensor is larger, and the images were taken at the same exposure and FOV, the larger sensor when re-sampled will beat the smaller sensor in SNR of the resulting image - almost irregardless of pixel density. Meaning, you will not find a 12, 18, 24, 36 or 43MP FF sensor that is 'beat' by an aps-c sensor of similar gen, of any MP - because more light went into those FF images in the first place.

But... that doesn't mean you can get an identical image out of them, though, because in order to get the same FoV and ISO, aperture and consequently DoF must vary.

In order to get the same FOV and exposure at the same time, DOF does need to vary, yes - an important aspect of equivalence that you need to be aware of. You have to be willing to 'accept' 1.3 stops less DOF to 'get' the better noise perf from the larger sensor. (I say 'accept', but in many cases it's not a negative to have less DOF, and in many cases it doesn't make any real difference to the shot - think shots where the whole subject is easily in the DOF in either case, or youre near hyperfocal.)

(Please tell me if that was coherent!)

Absolutely.

This one is a case where the sensors are not from the same wafer. They both have the same 16 megapixels, but with the D7000 being larger, the pixels are larger. The difference isn't from the size of the sensor, but in the size of the pixels on the sensor.

No, this is a common misconception. In a same-gen comparison, a larger sensor with greater pixel density than the smaller will still show less noise. It's about sensor area there, along with the mandated larger physical aperture at the same exposure for that FOV. Really, really, it is I can provide some external references if you wish for further reading - what I've found is it's very instructive to 'listen in' on a conversation where one of the conferees happens to be Eric Fossum, or Bob Newman, or Joseph Wizniewski - you know, the people who designed the hardware and software we use to create our delicious images

.

---------- Post added 03-17-16 at 12:56 PM ----------

The same arguments are presented over and over and over. The one thing that I am repeatedly failing to understand is Clackers' insistence that DXO's SCREEN tab is the relevant one. Sure, I understand his argument with respect to pixel/RAW noise, but in the end, one has to perform comparisons on the end result, i.e. a print or screen image. In either medium, the images from the different sensors must be produced to the same given size in order for the comparison to be meaningful. As I understand it, such a comparison is represented by the PRINT tab on the DXO graph.

Why is this even arguable?

I don't know. I just hope he can dial it and try to remain civil so this thread can continue to be useful. (But he's demonstrably wrong, stay tuned.)

[osv]

The same arguments are presented over and over and over. The one thing that I am repeatedly failing to understand is Clackers' insistence that DXO's SCREEN tab is the relevant one...

Why is this even arguable?

because it's clackers that's why i've got him on ignore.

even dxo itself tells you to use the print tab:

"Resolution-normalized
In order to compare DxOMark Camera Sensor measurements, you need to look at the “Print” mode (aka resolution-normalized) results. These can be directly compared across cameras with different resolutions. The overall DxOMark Sensor score is “Print” level. In the 3 more detailed scores you can choose between “Print” (default) and “Screen”. Just ignore the “Screen” setting." How fair is DxOMark ? - DxOMark

[jsherman999]

Ok so it's settled then. For all those who are happy to post the DXO graphs, you can all keep your K5ii bodies and all the rest of us will use K-1s and Sony A7rII bodies and we will never have to speak of this again.

Well, half-right, they would only be willing to consult the DxO 'screen' tab, because the screen tab tells them the K5 matches the FF K-1. Thou shalt not speak of the 'print tab', ever after.

[derekkite]

No you don't have to stop down to get the same dof on a ff vs apsc. Dof is a physical measurement of where in the scene in units of measure from the sensor plane the image is in focus, two numbers, near and far, with a third number being the focus plane, clarity decreasing gradually as you move from the focus plane.

That measurement will be very close on ff vs apsc, in fact the ff will be deeper.

If you want a percentage of the whole image to be in focus, say 1/3 from near to far, you have to stop down the ff because the dof at the same aperture is the same or close but a smaller proportion of the image.

So there are two things to know, the actual dof, and the proportion of the image that dof represents.

For example, if you are shooting a crowd from a platform and the crowd is 12 feet deep, to get the people in focus on an apsc and ff with the same lens would need the same aperture. The ff would have more scene nearer and further included and out of focus.

Do some numbers folks. There are numerous dof calculators out there.

---------- Post added 03-17-16 at 01:00 PM ----------

The screen tab is a one to one pixel noise reading of the image. The print tab is that number normalized to the size of the image from the sensor onto a fixed print size.

The d800e had close to a stop better noise characteristic than the k3 at higher ISO. But normalized, it is closer. And a k3 has a higher noise level than the k5iis, but normalized they are very close.

Both are of interest, neither in isolation tell the whole story, and which is predominant is very dependent on the specific shot.

I want two normalized curves showing not different ISO levels but same ISO at different crop percentages of the same sensor.

[jsherman999]

No you don't have to stop down to get the same dof on a ff vs apsc. ...

...
For example, if you are shooting a crowd from a platform and the crowd is 12 feet deep, to get the people in focus on an apsc and ff with the same lens would need the same aperture. The ff would have more scene nearer and further included and out of focus.

Keep in mind we're usually talking about the same FOV - meaning, not the same lens on both formats. If you shot the same lens on both formats, the FOV and image would be so radically different that any difference in DOF would be moot for most people.

Here:

• For an equivalent field of view, an APS-C crop sensor camera has at least 1.5x MORE depth of field that a 35mm full frame camera would have - when the focus distance is significantly less then the hyperfocal distance (but the 35mm format needs a lens with 1.5x the focal length to give the same view).

• Using the same lens on an APS-C crop sensor camera and a 35mm full frame body, the a APS-C crop sensor camera image has 1.5x LESS depth of field than the 35mm image would have (but they would be different images of course since the field of view would be different)

• If you use the same lens on a APS-C crop sensor camera and a 35mm full frame body and crop the full frame 35mm image to give the same view as the APS-C crop image, the depth of field is IDENTICAL (which assumes same display size and necessarily means the DOF changed from the originating FF image)

• If you use the same lens on a APS-C crop sensor camera and a 35mm full frame body, then shoot from different distances so that the view is the same, the APS-C crop sensor camera image will have 1.5x MORE DOF then the full frame image.

• Close to the hyperfocal distance, the APS-C crop sensor camera has a much more than 1.5x the DOF of a 35mm full frame camera. The hyperfocal distance of a APS-C crop sensor camera is 1.5x less than that of a 35mm full frame camera when used with a lens giving the same field of view.

[jsherman999]

The quote from this Link

"Print versus Screen mode

...The overall DxOMark Camera Sensor score is “Print” level only, which is fine. For the next level of detail a viewer gets to choose between Print and Screen. This is less fine: Screen is not normally useful for end users (it can be useful for debugging your own calculations). The lowest level of data is presented in “Screen” mode only, but is not labeled as such. I would prefer to see all data to be labeled Print/Screen or –better yet– Normal/100%. Normal would stress that this is what matters. And 100% is similar to pixel peeping: here you look at the noise at the 100% crop level and lose the overview of what it means at the image level."

I've always disliked the labels too, but because I encounter people who think one literally means "when I print" and the other means "when I look at it on my monitor (not at 100%)". And, I suspect those labels help lead indirectly to threads like this - if they were labeled "Normal" and "100%" as suggested above it would be more useful.


.

[npc]

The same arguments are presented over and over and over. The one thing that I am repeatedly failing to understand is Clackers' insistence that DXO's SCREEN tab is the relevant one. Sure, I understand his argument with respect to pixel/RAW noise, but in the end, one has to perform comparisons on the end result, i.e. a print or screen image. In either medium, the images from the different sensors must be produced to the same given size in order for the comparison to be meaningful. As I understand it, such a comparison is represented by the PRINT tab on the DXO graph.

Why is this even arguable?

Because of what happens before "the end result" and how the resolution of this "end result" you personally produce comares to what DXO uses as normalized size for their "Print" measurements.

For example I normally crop, sharpen, run some noise reduction algorithms , etc, etc, Then scale to the final size and maybe do this in steps with some additional fine sharpening in between.

What DXO does for the "Print" tab is just scale the image to something 8MP (not sure abut the mp, but 8 was given somewhere on their site as example so might very well be the value they are really using) using unknown to me scaling algorithm then measure the noise of that scaled image.

So if my process is different and my final output resolution is genarally different then their normalized values are just irrelevant.Normalization is only useful for comapison if everybody normalizes using the same process and to the same output.

What if in the future the norm becomes viewing images on a 24mpix screen for example (just to illustrate the point)? Then suddenly 16 mpix images whould actually need to be upscaled for this, amplifying their noise in the process.

[jsherman999]

For example I normally crop, sharpen, run some noise reduction algorithms , etc, etc, Then scale to the final size and maybe do this in steps with some additional fine sharpening in between.

Sure, but if you do all that to an image from camera A, and then do the same steps to the same scene but taken from camera B, and camera B has a stop on camera A according to DXOMark, your personal results will still likely be that much better from camera B. I think you'd have to really try something weird in PP to get comparative results that deviate from what the screen tab curve shows. I'm sure it's possible, but not likely in most people's PP routines. Your most likley path to that would to do something so destructive in PP that noise doesn't even matter anyway.

What DXO does for the "Print" tab is just scale the image to something 8MP ...
So if my final output resolution is genarally different then their normalized values are just irrelevant.

No they aren't, at least as far as resolution goes - meaning if you output to a normalized 4MP, or 12MP, or 24MP instead, the normalized values will shift by a constant. Which means, your curves would be the same at any MP output, DXO just happened to choose 8MP.

Here, from the previously shared link:

"...As can be seen, high-resolution sensors will gain more SNR, DR, TR and CS when reduced to a lower reference resolution. For DxOMark Sensor Overall Score and Metrics, we chose a reference resolution equal to 8 Megapixels, which is a bit less than a 12" x 8" print with a 300dpi printer. However, any other resolution can be chosen, as doing so only shifts the normalized values by a constant (because the reference resolution appears only as a logarithm in the formulas above). "

What if in the future the norm becomes viewing images on a 24mpix screen for example (just to illustrate the point)? Then suddenly 16 mpix images whould actually need to be upscaled for this, amplifying their noise in the process.

Yes, but this doesn't change the relative curves of any two cameras you're comparing for the reason DXO gives above, as they both would need to be upscaled to the same MP output target. Meaning, something like this will never be said: "camera A is 1 stop worse than Camera B at 8MP, but upsampled to 24MP Camera A suddenly matches it!"



.

[Ian Stuart Forsyth]

One of the stumbling points I had with understanding why one has to normalize the final image is this idea of image size, a ist ds image is no larger than a K3Ii image. You first have to understand that the image captured for the DS is 23.5 x 15.6 mm the same as the K3II 23.5 x 15.6 mm. Increasing the pixel count has no influence on the size of the image just as switching to a higher resolution film cannot change the size of the negative or slide. What is different is that the K3II has sampled the image at a far greater resolution and thus has an influence on how we see one of the contributors to noise, shot noise ( the random pattern of how light is distributed). With a lower resolution capture we are hiding that shot noise within the decreased resolution. With a higher resolution sensor the resolution at which we are scanning the image allows us to see how light naturally falls onto the sensor for the amount of total light falling on that sensor.


Now the reason why we have to compare normalized images is due to the fact that all images are scaled from the 23.5x 15.6 image (with in the same format) and the pixel has no influence other than the resolution captured at the time the shutter is pressed. Just as using a higher resolution film cannot change the size of the image captured, the only influence that higher res films has is the detail you will see when scaling (enlarging ) that 23.5 X 15.6 image to a given print size.

So when we look at the DXO and use the screen tab we are looking at the noise characteristics at different resolutions so yes we will see noise differently, what is important is the noise characteristics of the image contain at a common prints size.

[npc]

Well, on digital a pixel is a pixel. Could be of different size on the sensor but once sampled to data it does not have a size anymore just a value. So you just have data. It could be 16mpx from aps-c sensor or 24 from FF sensor or something else but it does not matter anymore since it only measures the noise based on the pixel values. As the output pixels are the same size if you use the same output media for both images then really more mpx = bigger source image size.
Then when you know your what is your raw noise + your resolution and this gives you some idea wether you need to denoise and how much without losing too much details, how much you can scale and how much potentially you can crop to get to your target output media resolution of 8mpix while keeping sufficient image quality. If you just look at a normalized value then these things are harder to tell as differences are averaged out when you scale both images down. Just to take it to the extreme if you scale the noisiest possible 4mpix image and the cleanest possible say 24 mpix image to a thumbnail of 16x16 px probably you can't measure any difference anymore off the thumbnais. Then try to scale the same to 8mpix ...

Anyway, It is pointess to compare 2 different mpx sensors just for noise without considering their resolution as well and normalizing the images to a third size does not tell the whole story.
For same mpix sensors regardless of their size there's no need to normalize since they are already "normalized".

[bxf]

It is pointess to compare 2 different mpx sensors just for noise without considering their resolution as well and normalizing the images to a third size does not tell the whole story

Then what does tell the whole story? Granted, my thinking is being driven just by common sense rather than technical knowledge, but that "third size" represents my end product - a print or screen image. What else should I be looking at?