# Crop Factor, Low Light and Aperture with m4/3 lenses? Part 2

Started 6 months ago | Discussions
 Like? 2
 Crop Factor, Low Light and Aperture with m4/3 lenses? Part 2 6 months ago

First: I hate this WYSIWYG editor - ruined one attempt totally. The 149 post limit (or 150?) is silly too. Come on DPR...

If your logic doesn't survive a platonic thought experiment, what sort of logic is it?

Your thought experiment was not only faulty, but also not relevant for the topic. The logic used was not really much more complex one 1+1=2.

What this does is that it perfectly downsizes the image drawn by the lens from 43mm diagonal to 21.6mm, thus you can duplicate the 43mm siized image diagonal perfectly as well.

You said earlier:

the more you need to enlarge the image, the better quality the image lens draws needs to be for one to get the result one wishes

What we have seen here is that it's not really the enlargement that is the issue.

It is precicely that, the enlargement factor.

Although we don't have any perfect physical speedboosters, we have some pretty good ones, and I think it's clear that the difference in resolution between two 20cm X 30cm prints generated with the same FF lens on an FF camera and on an m43 camera (with a sensor having the same resolution as the FF camera) using a good speedbooster will be much smaller than the difference suggested by the 69 and 267 times required enlargement factors you mentioned, or do you disagree?

You're right for most cases of course - I never claimed anything else.

However the problem is that this thought experiment has no relevancy for this issue: when you use the booster you downsize the image from 43mm diagonal to 21.6mm, thus you use the original lens as intended, all of it's lens area including all of the original image circle. The situation is quite different when you only use the center part of the image circle and enlarge that.

Now, let's talk about two different lenses - one designed to project an FF image circle, the other a smaller m43 image circle. Earlier, you said:

Regardless, there is a limit how good a lens can be, thus a larger format in principle has a higher ceiling for image quality in this regard as well.

You have not explained here what physical principle connects the first half of your sentence with the second half, especially taking into account two lenses designed to project different sized image circles?

All physical lenses have aberrations, do they not? The more these aberrations are enlarged, the lower the resulting image quality will be, thus there is need for better quality lenses when the enlargement is large. Also this is usually the case. I am quite certain the average m43 lens draws more linepairs per millimater on image plane than the average full frame lens because it needs to do that to be competetive.

How good a lens for different formats can be? There are many variables involved, some beneficial to small, some to the big - smaller elemets are easier to make close to perfect, on the other hand tolerances for larger formats are much larger, not just for the optical elements, but also for their mechanical positioning. Can I give you precise and simple answer how much smaller image circle influences which points in lens design and manufacturing, no I can not. But it is not just a reasonable statement due to the significant differences in enlargements, manufacturing tolerances and so on. I think there was a topic on this issue in the Techology/Science forum recently with much more qualified people than me involved.

Additionally there is the issue of diffraction. There is an absolute limit for image quality when it comes to resolution and the larger the format, the higher the ceiling is.

Without the perfect idealized speedbooster you can not enlarge from 21.6mm diagonal to 43mmm diagonal without the imperfections of the lens becoming more visible.

While this is true, the thought experiment is a simple illustration of the fact that we must not forget to take into account the differences in image circle size each lens is designed to project.

You're expanding the issue. All I claimed is that you need to have a better lens in a smaller format to get the same image quality. I see no real point in going to the direction you're driving as it is beside the point. I think that would be overly complex and fruitless.

A question: What is the mathematical relationship between sensor size and how much more visible the imperfections of a lens become?

If we use the very same lens, then exactly by the the same factor the crop factor is. Seriously, what else could it be? a(crop)=a(native)*cropfactor.

There must be a mathematical relationship, after all, if enlargement is the actual cause of the differences in lens performance w/r/t different formats that we're discussing.

Indeed, and it was presented above. All the enlargement does is that it enlarges everything, both details and aberrations. There really is nothing more to it.

Complain
 Like? 1
 Re: Crop Factor, Low Light and Aperture with m4/3 lenses? Part 2 In reply to Truthiness, 6 months ago

Truthiness wrote:

Although we don't have any perfect physical speedboosters, we have some pretty good ones, and I think it's clear that the difference in resolution between two 20cm X 30cm prints generated with the same FF lens on an FF camera and on an m43 camera (with a sensor having the same resolution as the FF camera) using a good speedbooster will be much smaller than the difference suggested by the 69 and 267 times required enlargement factors you mentioned, or do you disagree?

You're right for most cases of course - I never claimed anything else.

However the problem is that this thought experiment has no relevancy for this issue: when you use the booster you downsize the image from 43mm diagonal to 21.6mm, thus you use the original lens as intended, all of it's lens area including all of the original image circle. The situation is quite different when you only use the center part of the image circle and enlarge that.

So, in other words, all we need to do is design all m4/3 lenses with FF image circles and speedboosters built in and your "inherent" advantage for FF magnification disappears. Hmmmm...

Since magnification is the root of an inherent and unavoidable evil according to your original posting in the other thread, what do you make of the greater magnification required to project the larger image circle onto a FF sensor compared to the lesser magnification required to project an image circle on a m4/3 sensor? Seems to me that the FF sensor is "inherently" disadvantaged due to the "stress" of having a larger image circle.  Hmmmm...

Complain
 Like? 1
 Re: Crop Factor, Low Light and Aperture with m4/3 lenses? Part 2 In reply to Truthiness, 6 months ago

The post of yours that began this exchange in the previous thread was:

http://www.dpreview.com/forums/post/53861976

"Additionally there is one inherit advantage larger formats have over smaller ones: enlargement factor. A large format image needs to be enlarged less than a medium format image, which needs to be enlarged less than FF image which needs to be enlarged less than APS-C image which needs to be enlarged less than m4/3 image which needs to be enlarged less than cell phone image and when less enlargement is needed, the lens is stressed less, thus the lens can be of lesser quality to achieve the same image quality."

Exactly what do you mean by "stressed less" and "lesser quality" here? And in what way does the "lesser quality" actually translate into an "inherent advantage of larger formats"?

Anders W's gear list:Anders W's gear list
Panasonic Lumix DMC-G1 Olympus OM-D E-M5 Olympus E-M1 Panasonic Lumix G Vario 14-45mm F3.5-5.6 ASPH OIS Panasonic Lumix G Vario 7-14mm F4 ASPH +21 more
Complain
 Like? 1
 Reply to Just another Canon shooter In reply to Truthiness, 6 months ago

Just another Canon shooter wrote:

Anders W wrote:

Noise can't be "blurred". It's not affected by the optical resolution. The E-M1 shows less noise.

Of course it can, it is called NR,

NR blurs the detail. It reduces the noise.

It reduces the noise by blurring it.

A matter of semantics. My assumption was that no NR is applied in any of these cases.

The noise in the deep blue has very non-Poisson structure, as simple as that.

Can you please specify mathematically what you mean by a non-Poisson structure and how you can see or measure the deviation from Poisson in this case?

which can be done after the demosaicing, or incorporated in the demoasicing algorithm. Shot noise is of Poisson type, even on a Bayer sensor. If we make it B&W for simplicity, it has a well defined structure and spectrum. Missing high frequencies with the low frequencies present is an evidence of blurring, or NR, or call it whatever you want.

So are you saying that ACR secretly applies more NR to the E-M1 than to the other cameras although the NR sliders are set 0 in all cases?

You cannot separate demosaicing from NR. The old ACR process, for example, has "hidden NR" in it, by your logic.

So are you saying that ACR uses a different demosaicing algorithm for the E-M1 than for the other cameras?

Go to ISO 100. The deep blue square has more blurred borders compared to the others. Look at the 6D crop - no such thing.

The borders are a matter of resolution. Note what I already said about the E-M1 shot not being appropriate for judging resolution.

Resolution magically changes from one color to another?

It does not need to do that in order to blur the border.

It blurs one border but not another, while the camera next to it does not do that.

I have seen plenty of examples where different colors are differently blurred (especially when the border is black) due to properties of the lens rather than the sensor or the processing of the RAW file.

Lens aberrations cannot change the structure of the noise of a uniform target.

Right. That's one thing I had in mind when I said noise can't be blurred.

But processing can.

Sure. I was just explaining what I had in mind when I said noise can't be blurred.

I'd additionally appreciate a response to the following question included in my previous reply:

We had a poll here recently where two anonymous images were pitted against each other. One had a one stop advantage for SNR at 18 percent and the other had a one stop advantage for DR, but the poll participants didn't know anything about that in advance. The question asked was:

Which picture has higher image quality as far as noise is concerned?

What do you think the verdict of the poll participants was?

Anders W's gear list:Anders W's gear list
Panasonic Lumix DMC-G1 Olympus OM-D E-M5 Olympus E-M1 Panasonic Lumix G Vario 14-45mm F3.5-5.6 ASPH OIS Panasonic Lumix G Vario 7-14mm F4 ASPH +21 more
Complain
 Like? 1
 Re: Crop Factor, Low Light and Aperture with m4/3 lenses? Part 2 In reply to Truthiness, 6 months ago

Truthiness wrote:

All physical lenses have aberrations, do they not? The more these aberrations are enlarged, the lower the resulting image quality will be, thus there is need for better quality lenses when the enlargement is large. Also this is usually the case. I am quite certain the average m43 lens draws more linepairs per millimater on image plane than the average full frame lens because it needs to do that to be competetive.

How good a lens for different formats can be? There are many variables involved, some beneficial to small, some to the big - smaller elemets are easier to make close to perfect, on the other hand tolerances for larger formats are much larger, not just for the optical elements, but also for their mechanical positioning.

Also, for the same total light, the rays have to be bended more for a smaller lens, which creates additional problems. We still have to use the same glass with the same index of refraction though.

Can I give you precise and simple answer how much smaller image circle influences which points in lens design and manufacturing, no I can not. But it is not just a reasonable statement due to the significant differences in enlargements, manufacturing tolerances and so on. I think there was a topic on this issue in the Techology/Science forum recently with much more qualified people than me involved.

+1. Whatever the reason, higher resolution with larger sensors is a fact.

Just another Canon shooter's gear list:Just another Canon shooter's gear list
Canon EOS 5D Mark II Canon EF 15mm f/2.8 Fisheye Canon EF 35mm f/1.4L USM Canon EF 50mm f/1.2L USM Canon EF 135mm f/2.0L USM +4 more
Complain
 Like?
 Re: Reply to Just another Canon shooter In reply to Anders W, 6 months ago

Anders W wrote:

Just another Canon shooter wrote:

Anders W wrote:

Noise can't be "blurred". It's not affected by the optical resolution. The E-M1 shows less noise.

Of course it can, it is called NR,

NR blurs the detail. It reduces the noise.

It reduces the noise by blurring it.

A matter of semantics. My assumption was that no NR is applied in any of these cases.

I explained this already, twice. A less aggressive demosaicing can be thought of as NR, if you wish.

The noise in the deep blue has very non-Poisson structure, as simple as that.

Can you please specify mathematically what you mean by a non-Poisson structure and how you can see or measure the deviation from Poisson in this case?

Since images are taken for us to view them, I mostly trust my eyes - it looks like a NR reduced image. Non Poisson is just that - non Poisson. How to test for it - there many tests. The simplest one would be to frop a few squares and use ImageJ to compute the power spectrum - uniformity means Poisson, more or less; lower high frequencies means NR.

which can be done after the demosaicing, or incorporated in the demoasicing algorithm. Shot noise is of Poisson type, even on a Bayer sensor. If we make it B&W for simplicity, it has a well defined structure and spectrum. Missing high frequencies with the low frequencies present is an evidence of blurring, or NR, or call it whatever you want.

So are you saying that ACR secretly applies more NR to the E-M1 than to the other cameras although the NR sliders are set 0 in all cases?

You cannot separate demosaicing from NR. The old ACR process, for example, has "hidden NR" in it, by your logic.

So are you saying that ACR uses a different demosaicing algorithm for the E-M1 than for the other cameras?

Of course. With different parameters, at least, and giving different results for different colors.

Go to ISO 100. The deep blue square has more blurred borders compared to the others. Look at the 6D crop - no such thing.

The borders are a matter of resolution. Note what I already said about the E-M1 shot not being appropriate for judging resolution.

Resolution magically changes from one color to another?

It does not need to do that in order to blur the border.

It blurs one border but not another, while the camera next to it does not do that.

I have seen plenty of examples where different colors are differently blurred (especially when the border is black) due to properties of the lens rather than the sensor or the processing of the RAW file.

Combine that with the weird low frequency noise inside the blue square, and the lens gets eliminated as a factor.

But processing can.

Sure. I was just explaining what I had in mind when I said noise can't be blurred.

I'd additionally appreciate a response to the following question included in my previous reply:

We had a poll here recently where two anonymous images were pitted against each other. One had a one stop advantage for SNR at 18 percent and the other had a one stop advantage for DR, but the poll participants didn't know anything about that in advance. The question asked was:

Which picture has higher image quality as far as noise is concerned?

What do you think the verdict of the poll participants was?

Without seeing the experiment, I cannot answer. Was it a bright, a dark image, how processed, etc. And I do not really want to get in another long exchange.

Just another Canon shooter's gear list:Just another Canon shooter's gear list
Canon EOS 5D Mark II Canon EF 15mm f/2.8 Fisheye Canon EF 35mm f/1.4L USM Canon EF 50mm f/1.2L USM Canon EF 135mm f/2.0L USM +4 more
Complain
 Like?
 Re: Reply to Just another Canon shooter In reply to Just another Canon shooter, 6 months ago

Just another Canon shooter wrote:

Anders W wrote:

Just another Canon shooter wrote:

Anders W wrote:

Noise can't be "blurred". It's not affected by the optical resolution. The E-M1 shows less noise.

Of course it can, it is called NR,

NR blurs the detail. It reduces the noise.

It reduces the noise by blurring it.

A matter of semantics. My assumption was that no NR is applied in any of these cases.

I explained this already, twice. A less aggressive demosaicing can be thought of as NR, if you wish.

Sure. I also simply explained what I had in mind and not. My assumption was that the images are processed in effectively the same way as far as NR is concerned.

The noise in the deep blue has very non-Poisson structure, as simple as that.

Can you please specify mathematically what you mean by a non-Poisson structure and how you can see or measure the deviation from Poisson in this case?

Since images are taken for us to view them, I mostly trust my eyes - it looks like a NR reduced image. Non Poisson is just that - non Poisson. How to test for it - there many tests. The simplest one would be to frop a few squares and use ImageJ to compute the power spectrum - uniformity means Poisson, more or less; lower high frequencies means NR.

And did you perform any such analysis, for example for that dark blue square we are talking about, with the E-M1 at ISO 1600 and the 6D at ISO 6400, as exemplified by you here (with the link borrowed from your post here)?

which can be done after the demosaicing, or incorporated in the demoasicing algorithm. Shot noise is of Poisson type, even on a Bayer sensor. If we make it B&W for simplicity, it has a well defined structure and spectrum. Missing high frequencies with the low frequencies present is an evidence of blurring, or NR, or call it whatever you want.

So are you saying that ACR secretly applies more NR to the E-M1 than to the other cameras although the NR sliders are set 0 in all cases?

You cannot separate demosaicing from NR. The old ACR process, for example, has "hidden NR" in it, by your logic.

So are you saying that ACR uses a different demosaicing algorithm for the E-M1 than for the other cameras?

Of course. With different parameters, at least, and giving different results for different colors.

Specifically in a manner that affects the amount of NR effectively applied? Of course the color profiles used are not the same.

Go to ISO 100. The deep blue square has more blurred borders compared to the others. Look at the 6D crop - no such thing.

The borders are a matter of resolution. Note what I already said about the E-M1 shot not being appropriate for judging resolution.

Resolution magically changes from one color to another?

It does not need to do that in order to blur the border.

It blurs one border but not another, while the camera next to it does not do that.

I have seen plenty of examples where different colors are differently blurred (especially when the border is black) due to properties of the lens rather than the sensor or the processing of the RAW file.

Combine that with the weird low frequency noise inside the blue square, and the lens gets eliminated as a factor.

But processing can.

Sure. I was just explaining what I had in mind when I said noise can't be blurred.

I'd additionally appreciate a response to the following question included in my previous reply:

We had a poll here recently where two anonymous images were pitted against each other. One had a one stop advantage for SNR at 18 percent and the other had a one stop advantage for DR, but the poll participants didn't know anything about that in advance. The question asked was:

Which picture has higher image quality as far as noise is concerned?

What do you think the verdict of the poll participants was?

Without seeing the experiment, I cannot answer. Was it a bright, a dark image, how processed, etc.

It was the old DPR studio scene and the images were processed identically by means of DCRAW with everything at default.

And I do not really want to get in another long exchange.

The question is obviously relevant to the exchange we are already having.

Anders W's gear list:Anders W's gear list
Panasonic Lumix DMC-G1 Olympus OM-D E-M5 Olympus E-M1 Panasonic Lumix G Vario 14-45mm F3.5-5.6 ASPH OIS Panasonic Lumix G Vario 7-14mm F4 ASPH +21 more
Complain
 Like?
 Re: Crop Factor, Low Light and Aperture with m4/3 lenses? Part 2 In reply to knickerhawk, 6 months ago

knickerhawk wrote:

Truthiness wrote:

Although we don't have any perfect physical speedboosters, we have some pretty good ones, and I think it's clear that the difference in resolution between two 20cm X 30cm prints generated with the same FF lens on an FF camera and on an m43 camera (with a sensor having the same resolution as the FF camera) using a good speedbooster will be much smaller than the difference suggested by the 69 and 267 times required enlargement factors you mentioned, or do you disagree?

You're right for most cases of course - I never claimed anything else.

However the problem is that this thought experiment has no relevancy for this issue: when you use the booster you downsize the image from 43mm diagonal to 21.6mm, thus you use the original lens as intended, all of it's lens area including all of the original image circle. The situation is quite different when you only use the center part of the image circle and enlarge that.

So, in other words, all we need to do is design all m4/3 lenses with FF image circles and speedboosters built in and your "inherent" advantage for FF magnification disappears. Hmmmm...

Since magnification is the root of an inherent and unavoidable evil according to your original posting in the other thread, what do you make of the greater magnification required to project the larger image circle onto a FF sensor compared to the lesser magnification required to project an image circle on a m4/3 sensor? Seems to me that the FF sensor is "inherently" disadvantaged due to the "stress" of having a larger image circle. Hmmmm...

I kinda like these questions, knicker.

Anders W's gear list:Anders W's gear list
Panasonic Lumix DMC-G1 Olympus OM-D E-M5 Olympus E-M1 Panasonic Lumix G Vario 14-45mm F3.5-5.6 ASPH OIS Panasonic Lumix G Vario 7-14mm F4 ASPH +21 more
Complain
 Like?
 Anders, do you have the cameras and results again? In reply to Anders W, 6 months ago

I remember we all picked the lower res one as the better image in terms of noise, but I'd like the conclusion you made if you have it.

Bhima78's gear list:Bhima78's gear list
Olympus OM-D E-M10 Panasonic Leica D Summilux Asph 25mm F1.4 Sigma 60mm F2.8 DN | Art Olympus 12-40mm F2.8 Olympus M.Zuiko Digital 14-42mm 1:3.5-5.6 II R
Complain
 Like? 2
 Re: Anders, do you have the cameras and results again? In reply to Bhima78, 6 months ago

Bhima78 wrote:

I remember we all picked the lower res one as the better image in terms of noise, but I'd like the conclusion you made if you have it.

They were essentially the same res so that differences in this regard wouldn't complicate things. The poll was about signal-noise performance in the highlights and midtones (SNR 18%) versus signal-noise performance in the darker tones (DR). The latter was perceived as the more important factor by the large majority. Here's the poll report:

http://www.dpreview.com/forums/post/53675940

Anders W's gear list:Anders W's gear list
Panasonic Lumix DMC-G1 Olympus OM-D E-M5 Olympus E-M1 Panasonic Lumix G Vario 14-45mm F3.5-5.6 ASPH OIS Panasonic Lumix G Vario 7-14mm F4 ASPH +21 more
Complain
 Like?
 Re: Reply to Just another Canon shooter In reply to Anders W, 6 months ago

Anders W wrote:

Sure. I also simply explained what I had in mind and not. My assumption was that the images are processed in effectively the same way as far as NR is concerned.

Here are four crops. The smaller ones are the Oly, of course. The top row is processed with DXO Optics Pro, and the bottom one - with LR (copied from the screen of the link I posted). I equalized the overall brightness but that particular color appears darker in DXO.

Can you spot the NR?

6D (left) and Oly (right). Top: DXO. Bottom: ACR.

Clearly, LR applies more NR (or whatever you want to call it) to that color to the Oly than DXO. There is certain grain in the three crops missing from the bottom right one. Other colors are processed differently but I do not want to waste my time to post similar crops.

Just another Canon shooter's gear list:Just another Canon shooter's gear list
Canon EOS 5D Mark II Canon EF 15mm f/2.8 Fisheye Canon EF 35mm f/1.4L USM Canon EF 50mm f/1.2L USM Canon EF 135mm f/2.0L USM +4 more
Complain
 Like?
 Re: Crop Factor, Low Light and Aperture with m4/3 lenses? Part 2 In reply to Truthiness, 6 months ago

Just thinking out loud here.

Suppose that we have a 50mm F4 lens for a 35mm full frame, and a 25mm F2 lens for M43. These two lenses have equivalent focal length and aperture considering their designated sensors.

Now lets consider lens construction with the same front focal length and rear focal length (leave those complications for later).

The front element of the lens is the entrance pupil, it cannot be smaller than the effective maximum aperture. For the 50mm F4 lens the minimum front element diameter is 12.5mm, and it is the same for the 25mm F2 lens. (The actual element is probably slightly bigger than this)

The same consideration holds for the rear element of the lens.

This result seems to hold for my Panny 20mm F1.7 lens, for my FD50mm F1.4 lens the rear element is a little smaller - likely implying a slightly shortened rear focal length.

Now if the object plane has line pairs on it, these line pairs are shown on the M43 image plane as half the size than on the 35mm FF image plane.

Where does the image blur of the line pairs come from? The blur comes from not all the rays of a point source of the object focusing on the same point of the image on the sensor - some are focused in front of the sensor, some behind, some to the side.

Very crudely, rays which pass through an acceptable circle of confusion on the sensor are seen as sharp, those which are outside add blur.

Assuming the end images on computer screen or photographic paper are the same size, viewed from the same distance, the acceptable CoC on the M43 sensor is half the diameter of the acceptable CoC on the 35mm full frame sensor.

Now notice the the depth of focus around the M43 sensor is narrower than the depth of focus around the 35mm FF sensor, first because the CoC is half the diameter, but second because the rear focal length of the lens is half, and thus the rays from the lens come at the sensor over a cone which is twice as wide - twice the angle.

So, in order to maintain the same image sharpness, the depth of focus of the M43 lens must be nominally 4x narrower than that of the equivalent 35mm FF lens, although over 1/4 of the image area - but with the same FOV of the object plane.

I'm searching for way to compare the precsion required in the lens design and manufacture, maybe its this: - for a particular level of sharpness , the ratio of the depth of focus required to the image diagonal is a factor of two smaller for M43 than it is for 35mm FF.

-- hide signature --

Eric

Complain
 Like? 2
 Re: Reply to Just another Canon shooter In reply to Just another Canon shooter, 6 months ago

Just another Canon shooter wrote:

Anders W wrote:

Sure. I also simply explained what I had in mind and not. My assumption was that the images are processed in effectively the same way as far as NR is concerned.

Here are four crops. The smaller ones are the Oly, of course. The top row is processed with DXO Optics Pro, and the bottom one - with LR (copied from the screen of the link I posted). I equalized the overall brightness but that particular color appears darker in DXO.

Can you spot the NR?

6D (left) and Oly (right). Top: DXO. Bottom: ACR.

Clearly, LR applies more NR (or whatever you want to call it) to that color to the Oly than DXO. There is certain grain in the three crops missing from the bottom right one. Other colors are processed differently but I do not want to waste my time to post similar crops.

Here are Std. Dev. measurements for the four patches:

DXO 6D: 45.32

DXO Oly: 43.22

LR 6D: 40.55

LR Oly: 42.93

Lower Std. Dev. in LR for both cameras and the differences between DXO and LR is GREATER for the 6D than it is for the Oly.  Hmmmmm...

Complain
 Like?
 Re: Reply to Just another Canon shooter In reply to Just another Canon shooter, 6 months ago

Just another Canon shooter wrote:

Anders W wrote:

Just another Canon shooter wrote:

Anders W wrote:

Noise can't be "blurred". It's not affected by the optical resolution. The E-M1 shows less noise.

Of course it can, it is called NR,

NR blurs the detail. It reduces the noise.

It reduces the noise by blurring it.

A matter of semantics. My assumption was that no NR is applied in any of these cases.

I explained this already, twice. A less aggressive demosaicing can be thought of as NR, if you wish.

The noise in the deep blue has very non-Poisson structure, as simple as that.

Can you please specify mathematically what you mean by a non-Poisson structure and how you can see or measure the deviation from Poisson in this case?

Since images are taken for us to view them, I mostly trust my eyes - it looks like a NR reduced image. Non Poisson is just that - non Poisson. How to test for it - there many tests. The

Do you mean this

http://en.wikipedia.org/wiki/Poisson_distribution

simplest one would be to frop a few squares and use ImageJ to compute the power spectrum - uniformity means Poisson, more or less; lower high frequencies means NR.

which can be done after the demosaicing, or incorporated in the demoasicing algorithm. Shot noise is of Poisson type, even on a Bayer sensor. If we make it B&W for simplicity, it has a well defined structure and spectrum. Missing high frequencies with the low frequencies present is an evidence of blurring, or NR, or call it whatever you want.

So are you saying that ACR secretly applies more NR to the E-M1 than to the other cameras although the NR sliders are set 0 in all cases?

You cannot separate demosaicing from NR. The old ACR process, for example, has "hidden NR" in it, by your logic.

So are you saying that ACR uses a different demosaicing algorithm for the E-M1 than for the other cameras?

Of course. With different parameters, at least, and giving different results for different colors.

Go to ISO 100. The deep blue square has more blurred borders compared to the others. Look at the 6D crop - no such thing.

The borders are a matter of resolution. Note what I already said about the E-M1 shot not being appropriate for judging resolution.

Resolution magically changes from one color to another?

It does not need to do that in order to blur the border.

It blurs one border but not another, while the camera next to it does not do that.

I have seen plenty of examples where different colors are differently blurred (especially when the border is black) due to properties of the lens rather than the sensor or the processing of the RAW file.

Combine that with the weird low frequency noise inside the blue square, and the lens gets eliminated as a factor.

But processing can.

Sure. I was just explaining what I had in mind when I said noise can't be blurred.

I'd additionally appreciate a response to the following question included in my previous reply:

We had a poll here recently where two anonymous images were pitted against each other. One had a one stop advantage for SNR at 18 percent and the other had a one stop advantage for DR, but the poll participants didn't know anything about that in advance. The question asked was:

Which picture has higher image quality as far as noise is concerned?

What do you think the verdict of the poll participants was?

Without seeing the experiment, I cannot answer. Was it a bright, a dark image, how processed, etc. And I do not really want to get in another long exchange.

-- hide signature --

::> I make spelling mistakes. May Dog forgive me for this.

Complain
 Like? 3
 Perceived as more important? In reply to Anders W, 6 months ago

Anders W wrote:

They were essentially the same res so that differences in this regard wouldn't complicate things. The poll was about signal-noise performance in the highlights and midtones (SNR 18%) versus signal-noise performance in the darker tones (DR). The latter was perceived as the more important factor by the large majority. Here's the poll report:

http://www.dpreview.com/forums/post/53675940

Just last week there was a thread about ND Grads vs Bracketing where most responder consistently (and persistantly) agreed they would not use filters for landscapes, but would prefer bracketing instead. Yet noone could post images (on that very same thread) as nearly as good as the poster who prefers to use filters almost exclusively,

http://www.dpreview.com/forums/post/53827411

Most on this forum will insist they will stop the lens down to infinity when photographing macro (because it is an inherent advantage of the cropped formats), even the closeup and macro "specialist" himself once almost "taught" this,

http://www.dpreview.com/forums/post/52920712

Yet it is rather common not to stop the lens down past f/11 (f/5.6 on mFT) for most pleasing closeups compositions, and this has been demonstrated enough times on the forum as well.

The efficiency of the sensor is claimed to be present across the entire range

http://www.dpreview.com/forums/post/53862930

Yet we know at the most common level of (the perceived by us) light it is usually not there (SNR18%). Several people observed and pointed to it on the poll thread as well. Here was my post,

http://www.dpreview.com/forums/post/53659634

A shows to be less noisy in the dark areas, whereas B pulls slightly ahead in the mid-tones. At the computer display size image B looks more present to me (just by clicking back and forward), whereas A looks partially washed, and partially darker. Could be just a time of a day that made slight difference, I would not tell.

In essence it made no difference for any practical purposes whatsoever, yet someone who can not hold his camera straight knows all about it and how it's all supposed to mean something at the end. I mean, asking for dillydally entertainment, what other joys (other than going out and enjoying the day) would someone with the camera have. (Think twice before you think you know the answer).

-- hide signature --

- sergey

Sergey_Green's gear list:Sergey_Green's gear list
Samsung Galaxy Note 10.1
Complain
 Like? 3
 Re: Perceived as more important? In reply to Sergey_Green, 6 months ago

Sergey_Green wrote:

Just last week there was a thread about ND Grads vs Bracketing where most responder consistently (and persistantly) agreed they would not use filters for landscapes, but would prefer bracketing instead. Yet noone could post images (on that very same thread) as nearly as good as the poster who prefers to use filters almost exclusively,

http://www.dpreview.com/forums/post/53827411

While I agree that the images are nice, I doubt that the choice of filters over bracketing had anything to do with it.

Ulric's gear list:Ulric's gear list
Panasonic Lumix DMC-GF3 Olympus OM-D E-M5 Panasonic Lumix DMC-GM1 Panasonic Lumix G 20mm F1.7 ASPH Panasonic Lumix G 14mm F2.5 ASPH +6 more
Complain
 Like? 4
 Re: Crop Factor, Low Light and Aperture with m4/3 lenses? Part 2 In reply to knickerhawk, 6 months ago

knickerhawk wrote:

Truthiness wrote:

Although we don't have any perfect physical speedboosters, we have some pretty good ones, and I think it's clear that the difference in resolution between two 20cm X 30cm prints generated with the same FF lens on an FF camera and on an m43 camera (with a sensor having the same resolution as the FF camera) using a good speedbooster will be much smaller than the difference suggested by the 69 and 267 times required enlargement factors you mentioned, or do you disagree?

You're right for most cases of course - I never claimed anything else.

However the problem is that this thought experiment has no relevancy for this issue: when you use the booster you downsize the image from 43mm diagonal to 21.6mm, thus you use the original lens as intended, all of it's lens area including all of the original image circle. The situation is quite different when you only use the center part of the image circle and enlarge that.

So, in other words, all we need to do is design all m4/3 lenses with FF image circles and speedboosters built in and your "inherent" advantage for FF magnification disappears. Hmmmm...

This was the approach used by Olympus for their SHG zooms. Unfortunately, do it that way, all size advantage for the smaller sensor disappears.

In the end, sensor size doesn't matter very much, it is the size of the lens. The time when sensor size does matter is when common use-cases get too close to the f/0.5 limit of a lens.

-- hide signature --

Bob

Complain
 Like?

Re : assume the ISO is the same for a certain FF and MFT, then, for a particular scene if the FF can properly expose at 50mm FL, ISO 100, f/2.8 and 1/60", a MFT could have the same equivalent coverage (FOV) and same exposure at 25mm FL, ISO 100, f/2.8 and 1/60". The only difference is FOD?

Yes , that's it .. for the basics ( I assume you mean DOF , depth of field at the end ) ...

-- hide signature --

- Franka -

Complain
 Like?
 Re: Reply to Just another Canon shooter In reply to knickerhawk, 6 months ago

knickerhawk wrote:

DXO 6D: 45.32

DXO Oly: 43.22

LR 6D: 40.55

LR Oly: 42.93

Lower Std. Dev. in LR for both cameras and the differences between DXO and LR is GREATER for the 6D than it is for the Oly. Hmmmmm...

It should be, the crops are not normalized to the same area. To normalize, you need to shrink the 6D ones about 14.5%. Then they will look smaller (because dpreview frames for the same height, not the same diagonal). I will leave the calculations for you, and you may very well discover that the 6D wins.

Did you follow the discussion so far? This is the noisiest color for the 6D.

Just another Canon shooter's gear list:Just another Canon shooter's gear list
Canon EOS 5D Mark II Canon EF 15mm f/2.8 Fisheye Canon EF 35mm f/1.4L USM Canon EF 50mm f/1.2L USM Canon EF 135mm f/2.0L USM +4 more
Complain
 Like? 2
 Re: Perceived as more important? In reply to Sergey_Green, 6 months ago

Sergey_Green wrote:

Anders W wrote:

They were essentially the same res so that differences in this regard wouldn't complicate things. The poll was about signal-noise performance in the highlights and midtones (SNR 18%) versus signal-noise performance in the darker tones (DR). The latter was perceived as the more important factor by the large majority. Here's the poll report:

http://www.dpreview.com/forums/post/53675940

Just last week there was a thread about ND Grads vs Bracketing where most responder consistently (and persistantly) agreed they would not use filters for landscapes, but would prefer bracketing instead. Yet noone could post images (on that very same thread) as nearly as good as the poster who prefers to use filters almost exclusively,

http://www.dpreview.com/forums/post/53827411

Look who's talking. You don't even use GND filters yourself. Odd, isn't it, that he was never able to explain why/how an optical GND filter was superior to the digital solution? Since you established in that thread that you don't know anything about GND filters or merging+masking in PS, why don't you take one of goWex's classes and then take one of John Paul Caponigro's classes and report back to us on which of the two has a deeper understanding of the issue.

Most on this forum will insist they will stop the lens down to infinity when photographing macro (because it is an inherent advantage of the cropped formats), even the closeup and macro "specialist" himself once almost "taught" this,

http://www.dpreview.com/forums/post/52920712

Yet it is rather common not to stop the lens down past f/11 (f/5.6 on mFT) for most pleasing closeups compositions, and this has been demonstrated enough times on the forum as well.

Ironic, isn't it, that of the most recent (Summer 2014) close-ups you've posted to your site, the better ones are f/16 (one is even f/45?).  The ones at f/11 all exhibit problematic OOF issues on the flower-subject.  Oh well, live and learn...

The efficiency of the sensor is claimed to be present across the entire range

http://www.dpreview.com/forums/post/53862930

Yet we know at the most common level of (the perceived by us) light it is usually not there (SNR18%). Several people observed and pointed to it on the poll thread as well. Here was my post,

In essence it made no difference for any practical purposes whatsoever, yet someone who can not hold his camera straight knows all about it and how it's all supposed to mean something at the end. I mean, asking for dillydally entertainment, what other joys (other than going out and enjoying the day) would someone with the camera have. (Think twice before you think you know the answer).

And what joys do you have constantly hanging out in a forum for a format you don't even shoot? (Think twice before you think you know the answer.)

Complain