7D or 1DIV: better noise?

[arash_hazeghi]

Yes hi Arash,

When Roger equalized entrance pupil diameters between 7D and 1DIV, this ensures that the same # of photons was delivered to each pixel. If we go by F#, this does mean that the illuminance, or irradiance, or #photons/sq. micron at the focal plane is higher for the 7D than the 1DIV, but to get to # of photons, we have to multiply this #photons/sq.micron by each pixel's area, thus the 1DIV's higher pixel area compensates for this difference.

All Roger is saying is the following from radiative transfer:

P = L * A * a / (f^2) , where P is light power collected by a pixel (# photons in/sec) , A is the area of the entrance pupil, a is pixel area , f is focal length, and L is the radiance (luminance) or brightness of the object that we're taking a picture of. This takes the form of :
P = L * A * omega , where omega is a solid angle. Roger said that he is holding the number of pixels on the subject as constant, so he is holding (p/f)^2 constant (where p is the pixel pitch and a = p^2), which just says that p/f , which is the FOV that a pixel sees, is constant between the 7D and 1DIV. If that quantity is held constant, then the only quantity that determines P is A, the entrance pupil area, and Roger is holding that constant as well between the 2 cameras. So he is sending in the same amount of photons/sec, if using the same shutter speed, then the # photons incident is the same between the 2 pixels. Okay so this just leaves the QE, and from both of Roger's tests and from the Sensorgen data, these are very similar, giving a similar # of electrons collected in an exposure time. (Sensorgen data, btw, are done from curve fitting and data reduction of the Dxomark full SNR and measured ISO data).

Apart from read noise and PRNU, which are small effects at these ISOs, these 2 scenarios will give very similar SNRs as Roger has shown. Let's assume part of the picture has about 700 e-. For 7D at ISO 400, this would amount to about 1148 DN (using Roger's measured camera gain constant of 0.61 e-/DN). For 1DIV it would be about 1346 DN. Both of these are pretty similar "brightness" for the 14-bit system. (I use "brightness" in quotes because these are in units of DN).

Anyway, in short, Roger's tests showing the utility of etendue considerations, sound good to me.

Chris

Chris,
That's what I mean too, the total number of photons per unit area was not equal. The total number of photons per pixel was equal because 1D has larger pixels. But the point of having larger pixels is to collect more light, otherwise of course SNR is the same.

so I think identical illuminence should be used.

Remember Roger wants to compare 1D4 and 7D in identical field conditions. Here what he has proven is that Si absorption and cell QE is similar in both sensors, well of course it is :)


Am I missing something?

Thanks

[arash_hazeghi]

Roger,

the point of larger pixels of the 1D is to collect more light not the same as 7D. Why did you equalize the photons? when you measure SNR do you also assume the same number of photons per pixel as you did here? I don't think so

I still think the test in not right, you should have equalized illuminance or the number of photons per unit area, NOT the number of photons per pixel.

For a given shutter speed and f-number you will get the same photons/per area in both cameras thus 1D pixels will collect more photons...

Of course if you send the same number of photons a larger pixel has ZERO advantage over a smaller pixel and you are wasting it. It's a very strange method.

What was the rational behind your test? What field condition does it exactly represent?

thanks

[Alan Stankevitz]

The only way to do a fair comparison is to do a side-by-side test with both cameras, using the same subject from the same distance. How else can you possibly evaluate the merits of their image quality? (Excluding portrait photography, in which you move closer to frame your subject as they patiently wait for you.)

[Doug Brown]

Doug,
In your example, you changed the subject between the two cameras! That's not an equal comparison. To equalize the comparison, to image the same subject, and change lenses instead of distance.

It may not be a scientifically equal comparison, but it is a practical comparison. This is how I use my camera. This is how I choose my photographic subjects. Why would I slap on a converter if I could simply point the camera at a closer bird in flight or move closer to a stationary subject. Here's another hypothetical situation. I'm in Costa Rica photographing a Toucan with my Mark IV and a 500mm lens; I've done my composing in camera and have a perfect full-frame image. My Mark IV quits working so I grab my 7D backup body. My next step is to move the tripod away from the Toucan until I duplicate my original composition. I do not leave the tripod in the same location and crowd the bird in the frame. Now I have two compositionally identical frames taken with two different bodies. Why is it not valid to compare noise in the two frames? I understand that there are focal length limited scenarios, but those are the exception and not the rule in my photographic world. Based upon my camera usage style, isn't my test the best test to compare noise?

[Alan Stankevitz]

Doug,

The majority of the time, I do not have the luxury you speak of. I can see your point if you live in a world in which you are not focal-length-limited. Birds here in the Midwest are much too skittish. I also think you are dismissing other bird photographers who also do not live in your world. And yes, I use blinds, I use my car, etc.

If this were the case where we did not photograph in a focal-length-limited world, there would be no reason to own 500mm, 600mm, 800mm lenses nor teleconverters. More often than not, I'm looking for greater reach when photographing birds.

Alan

[Doug Brown]

The only way to do a fair comparison is to do a side-by-side test with both cameras, using the same subject from the same distance. How else can you possibly evaluate the merits of their image quality?

I'm just curious why that is the only way to do a fair comparison? What is it that's inherently wrong about the test that I outlined above? My test is about taking an identical frame with 2 different bodies and comparing noise. When you use a crop body do you compose your shots any differently than when you use a full frame body? I don't.

[Doug Brown]

I also think you are dismissing other bird photographers who also do not live in your world. And yes, I use blinds, I use my car, etc.

If this were the case where we did not photograph in a focal-length-limited world, there would be no reason to own 500mm, 600mm, 800mm lenses nor teleconverters. More often than not, I'm looking for greater reach when photographing birds.

Hi Alan. First of all, I clearly stated that my photographic world tends to not be focal-length limited. I wasn't being dismissive of other photographers. I also understand that bird photography requires lots of reach. But this thread is about relative noise levels. As long as the bare lens aperture I've selected is f/5.6 or smaller on my 500mm f/4 (which is most of the time), putting a 1.4x on my Mark IV incurs no noise penalty whatsoever. But it does neutralize the 'reach advantage' of the 7D. Reach and noise function independently from one another unless you are forced to bump your ISO in order to maintain an exposure.

How in a thread comparing the noise of a 7D to a Mark IV can you devise a test in which one camera is set to ISO 400 and the other to ISO 800? What is fair about that comparison? That test seems to have been designed to handicap the Mark IV by limiting the number of photons of light per pixel in order to match the number of photons per pixel on the 7D. But one of the major advantages of the Mark IV is that it is capable of collecting more light per pixel than the 7D due to its larger pixel size. It'd be like comparing the acceleration of a 12 cylinder Ferrari to that of a 6 cylinder Honda and disabling 6 of the Ferrari's cylinders in order to better match the Honda's engine size.

[Michael Gerald-Yamasaki]

Folks,

Greetings. This thread is quite amusing. FWIW...

- Objectively backwards deriving the differences between two sensors regarding noise characteristics from images is, IMO, impossible. Don't even try... you can't begin to equate all the variables, especially, since there are pre-RAW output secret processing sauces applied, let alone the differential treatment applied in distance to subject, cropping, ISO, sensor size that comes with "equating" the variables.

- Best guess is the collective opinion of expert users, where the collective output viewed is large, and not annecdotal. Meaning the "I shoot both. I shoot a lot. I think this one is better." is the best information you can get especially if there are a number of such opinions.

By that criteria, without even touching the 7D or Mark IV (I shoot Nikon), it appears the Mark IV is better for noise .

On that other point, perhaps one should say bird photography is largely about field skills, focal length changes the dynamic between field skills on your feet and field skills managing the focal length. There is no free lunch.

Cheers,

-Michael-

[Roger Clark]

Roger,

the point of larger pixels of the 1D is to collect more light not the same as 7D. Why did you equalize the photons? when you measure SNR do you also assume the same number of photons per pixel as you did here? I don't think so

Arash,
1st, I equalized the entrance pupil. If you don't start by feeding the camera the same amount of light, how can you conclude if it is noisier? There are no assumptions. It is a straight calculation of system throughput. Check Chris' math, or go to the wikipedia page on Etendue. This is not exotic stuff. It is exactly what we use when designing sensors for aircraft, spacecraft and telescopes. Physics works.

I still think the test in not right, you should have equalized illuminance or the number of photons per unit area, NOT the number of photons per pixel.

No. If you change the system to equalize illuminance, then you have changed parameters that make something unequal, including total light entering the system and depth of field.

Try this test: on a dark night, point your camera at the north star and do 30 second exposures at ISO 1600. Put something like a 50 mm lens on the camera at f/8. Then do it with your 500 mm lens at f/8. Which image records fainter stars. After all the illimunance by f/ratio is the same. But you will not get the same answer (number of faint stars).

For a given shutter speed and f-number you will get the same photons/per area in both cameras thus 1D pixels will collect more photons...

But if you freeze f/ratio, when you change focal length, you change the lens clear aperture diameter, which controls the amount of light. If you say keep the lens the same (e.g. 500 mm at f/8 on both cameras), you are then changing the image detail.

By your idea, just put on a 500 mm /f4 lens to do your bird photography. After all it has the same illuminance as a 500 f/4. See the fallacy?

Of course if you send the same number of photons a larger pixel has ZERO advantage over a smaller pixel and you are wasting it. It's a very strange method.

What was the rational behind your test? What field condition does it exactly represent?

thanks

The question was: does the 7D produce noisier images than the 1D4? To test that one needs to use the same conditions, not different incident amount of light, not different resolution on the subject, not different exposure times. In the field use is simple: smaller pixels are like having a built in TC. People on BPN have stated that want to move to a 1D camera to get AF at f/8. But with the small pixels of the 7D, it is already an advantage, so one doesn't need 2x TC and f/8 AF like that needed on a 1D when you need the reach. Or for example, say I have a 5D2 and want more reach. Should I buy a bigger lens? With the 7D one has 1.48 times more detail, so it is like having a 1.48x TC on the lens compared to the 5D2. If yu are in the field and have these two cameras, knowing the different reach that pixel size and focal length gives helps choose the right combination for a given situation. Then knowing Etendue helps one manage S/N in any given situation. Plenty of practical information. But one must get over f/ratios and ISO controls everything--it doesn't.

Roger

[Enrique Patino]

Opening both RAW files using Bridge, the subject on the 7D image seems a tad more underexposed, but not much (see numbers and histogram). I composed an image of both images after zeroing the blacks slider, leaving everything else the same. And here it is for your view. Top Image is 7D

As a practical exercise, let's assume we are in the field and have our cameras set to certain respective settings (according to Roger's respective settings for each camera system for his test) and a BIF appears out of nowhere. We do not have time to adjust setting and we fire away. These are the respective results.

Now, lets see how we can do with these underexposed images, including how much details we can pull out of the shadows... there is another thread under Digital Photography Workflow - processing RAW images exercise. What do people think of using these two RAW files provided by Roger as an on-the-side exercise and see what we can do? I think it would be interesting and educational (i.e. would add value to this thread, for me anyway )

[arash_hazeghi]

Roger,


I disagree. I believe you should equalize illuminance.

Please my answer my question regarding SNR. when you measure SNR do you assume same number of photons per pixel or not? yes or no? short answer

If you assume same number of photons per pixel then why do you get a higher SNR for the 1D in your own data? Why does your data scale with pixel size? Then all of your own data and analysis will be invalidated. Here I am attaching your own data.

How many photons did you assume for 1D pixel?
How many photons did you assume for 7D pixel?
Where these two numbers equal?

What you are saying is like I have a 10 gallon tank and a 5 gallon tank but I put only one gallon of gas in them to say their capacity is equal

[arash_hazeghi]

Here is how I photograph in the field

I use the 7D with 500mm lens f/8 1/1000sec ISO 800,

I swap the bodies, add a a TC but I don't change my exposure it is still f/8 1/1000sec ISO 800.

The aperture diameter is different but the number of photons per unit area is the same. The 1D has larger pixels so they collects more light and would have a larger SNR just as Roger's SNR data shows above.

This is how I would test :)

[Roger Clark]

Hi Alan. First of all, I clearly stated that my photographic world tends to not be focal-length limited. I wasn't being dismissive of other photographers. I also understand that bird photography requires lots of reach. But this thread is about relative noise levels. As long as the bare lens aperture I've selected is f/5.6 or smaller on my 500mm f/4 (which is most of the time), putting a 1.4x on my Mark IV incurs no noise penalty whatsoever.

Doug,
What happens when you add the TC? Your exposure time must be increased. If you didn't increase your exposure time, you would have less light per pixel so your S/N would drop. There definitely is a noise penalty.

But it does neutralize the 'reach advantage' of the 7D. Reach and noise function independently from one another unless you are forced to bump your ISO in order to maintain an exposure.

No. Reach and noise are intimately coupled and ISO is independent (mostly--see previous discussions). When you change ISO, it instructs the camera to change shutter speed and that is what changes the amount of light captured and determines the noise level. ISO does not change sensitivity or the amount of light collected by the sensor. To get more reach, you must increase focal length or chop the focal plane into =smaller pieces (smaller pixels)--no different.

How in a thread comparing the noise of a 7D to a Mark IV can you devise a test in which one camera is set to ISO 400 and the other to ISO 800?

Because ISO does NOT change the light captured. ISO is an independent gain. Set up an 18% gray target and set to manual and say meter so the histogram is on the right for ISO 3200. Not drop ISO by factors of two down to say ISO 200. Analyze the S/N on each. You will find a very weak dependence: at the high end, there will be no change is S/N and at lower ISOs, S/N will drop slightly due to the A/D and downstream electronics noise.

What is fair about that comparison? That test seems to have been designed to handicap the Mark IV by limiting the number of photons of light per pixel in order to match the number of photons per pixel on the 7D. But one of the major advantages of the Mark IV is that it is capable of collecting more light per pixel than the 7D due to its larger pixel size.

Now, the test equalized the light per pixel, the exposure time, the depth of field, the detail on the subject. What you and Arash propose is to make one or more of these conditions unequal.

It'd be like comparing the acceleration of a 12 cylinder Ferrari to that of a 6 cylinder Honda and disabling 6 of the Ferrari's cylinders in order to better match the Honda's engine size.

Nothing was disabled in the test. What you and Arash are proposing is limiting and inhibiting one camera to be able to perform at its fullest (the 7D). For example, if both cameras were at the same f/8 f/ratio, the 300 mm lens on the 7D would have an aperture of 37.5 mm but the 420 mm on the 1D4 would be 52.5 mm delivering 1.96 times more light. You see, it is not pixel size alone that collects more light in the situations you discuss, it is the combination of larger diameter lens and pixel size.

When one scales from a crop sensor camera to full frame and you change the lens to equalize the field of view, and you keep the f/ratio constant, the lens diameter is increasing, collecting more light and that enables the larger pixels to collect all that light. It is like a rain storm: if it rains harder, one needs a larger bucket to collect all that rain before spilling over the top. With cameras, it is the lens that delivers the light (bigger lens is like a stronger rain storm), the larger pixel only enables collection of more light before filling up. So larger pixels need a larger lens diameter to collect the light faster. It is the combination of lens diameter that collects the light, focal length that spreads out the light, and and pixel size that chops up the light into image detail that play the important factors in image detail and S/N.

Roger

[arash_hazeghi]

Doug,
What happens when you add the TC? Your exposure time must be increased. If you didn't increase your exposure time, you would have less light per pixel so your S/N would drop. There definitely is a noise penalty.

Roger

No, No! he is saying that he shoots at f/5.6 in both cases, the light intensity W/cm^2 or photons per um^2 is the SAME. The larger pixel then collects more light :)

[Luis Villablanca]

No, No! he is saying that he shoots at f/5.6 in both cases, the light intensity W/cm^2 or photons per um^2 is the SAME. The larger pixel then collects more light :)

I tend to agree with this. If my options were between a 7D + 500/4 @ f/5.6 and a 1D4 + 500/4 + 1.4x TC @ f/5.6, then I'd choose the 1D4.

However, If my options are between a 7D + 800/5.6 @ f/8 and 1D4 + 800/5.6 + 1.4x TC @ f/8 (which is something I face from time to time),
I'd probably choose the 7D, as it will be faster, with full AF, and the image would probably be sharper.
Also, if I stopped down the 1D (say to f/11), then I'd be adding more noise since I'd have to bump the ISO to get the same exposure as the 7D.

[Enrique Patino]

Two years can make a difference in (almost) the same subject

http://www.birdphotographers.net/for...ng-is-required

[arash_hazeghi]

Two years can make a difference in (almost) the same subject

http://www.birdphotographers.net/for...ng-is-required

I don't see anything different in my arguments at least. Gush 2.5 years is a long time.

Here is what Roger said back then

"I agree with Arash, and good description on the resampling issues with noise. One thing that strikes me on these test images is how close the resolution is between the 5DII and 7D. The 7D images also show less contrast, which is also predicted by the modulation transfer function. The reduction in contrast is an indication of loss of detail.

If the test were redone at higher ISO, the 7D would be degraded more. An a test of text does not show the fine detail that a feather test would show.

Roger "

Well two years ago Roger and I agreed with everything. But apparently I have done something that has made Roger angry and he doesn't like me anymore. I don't really know what it is though so maybe I can resolve the misunderstanding because I respect Roger and the effort he has put in the measurements and his website ;) I like to leave a positive impression not to make enemies, at the end of the day it's just a camera just take photos and be happy :)

[Alan Stankevitz]

Two years can make a difference in (almost) the same subject

http://www.birdphotographers.net/for...ng-is-required

Hey! I remember that thread now! Yikes! Two years ago? Since that time, I did buy the Mark IV and have owned it and two 7Ds. I still believe my test is a valid test. Cropping is a normal part of the workflow and if you were to take the same identical shot with the two cameras from the same distance to the subject, you will have to crop heavier with the Mark IV. The heavier an image is cropped, the more the in inherent noise will be revealed. This does have an effect on the final post-processed image.

Just like two years ago, I am anxious to see RAW images from the 1DX. Being a full-frame sensor, it will require heavier cropping than the Mark IV. How will the images compare once post-processed? How will the resolution hold up? It will be an interesting comparison.

Alan

[Daniel Cadieux]

at the end of the day it's just a camera just take photos and be happy :)

Man, this thread has been a whirlwind for sure and quite informative. Lots of amazing analysis and graphs and charts and all....but that simple statement above is the best one of them all!

[Dan Brown]

So, Please, Roger and Arash, come to an agreement! And then have Doug explain it!

[Roger Clark]

I don't see anything different in my arguments at least. Gush 2.5 years is a long time.

Here is what Roger said back then

"I agree with Arash, and good description on the resampling issues with noise. One thing that strikes me on these test images is how close the resolution is between the 5DII and 7D. The 7D images also show less contrast, which is also predicted by the modulation transfer function. The reduction in contrast is an indication of loss of detail.

If the test were redone at higher ISO, the 7D would be degraded more. An a test of text does not show the fine detail that a feather test would show.

Roger "

Well two years ago Roger and I agreed with everything. But apparently I have done something that has made Roger angry and he doesn't like me anymore. I don't really know what it is though so maybe I can resolve the misunderstanding because I respect Roger and the effort he has put in the measurements and his website ;) I like to leave a positive impression not to make enemies, at the end of the day it's just a camera just take photos and be happy :)

Arash,

"I have done something that has made Roger angry" is not at all the case; I am not angry, but frustrated that you only see one position.

The above-referenced thread was about the 7D before much was known about the 1DIV, and if you look further down the thread you will find other statements where I said small pixels like 5 microns would make an ideal camera. But given facts and a strong argument, I'll see the way, admit I'm wrong and learn and move forward. Like phase angle, where the varied images on BPN has taught me to realize how important phase angle is in image quality. Similarly, as new cameras have come out with varying sensor sizes and pixel sizes, I have realized the importance of Etendue in photography and telephoto reach in particular. I have used Etendue for decades in spacecraft and aircraft system remote sensing science, but now find it important in telephoto photography where reach is important. Thus, my views are more refined than they were 2 years ago in this area.

In fact two years ago I did not own a 7D. I first bought a 1D4 and later after using a friend's 7D and applying Etendue seeing how useful the 7D small pixels are, I bought a 7D. Over the last few years my sensor performance page has evolved with more on pixel size effects in focal length limited situations.

Arash, you advocate using the illuminance in the focal plane, so the equivalent metric on the focal plane is the photons per square micron (or mm or inch). Why not reference that plot from my web page? See Figure 10 at:
http://www.clarkvision.com/articles/...x.html#DENSITY
There the 7D plots above the 1D4.

What has been discussed with no change in views are two camps: Those that change position or light per pixel for the test, and those that fix the distance and hold as much as possible constant. This boils down to image quality in 2 different situations. Doug's position is given by:

Full Sensor Apparent Image Quality (FSAIQ) which is shown here:
http://www.clarkvision.com/articles/...dex.html#FSAIQ
Here we see the 1D4 plots well above the 7D, mainly because the sensor is larger and to fill the frame one is changing the test to favor the larger sensor. This is also commonly the
case for landscape photography: larger sensor wins. I acknowledge Doug's position is valid for the case he photographs in and I have already stated so in this thread.

And the position I and Alan were discussing is:
Focal Length Limited Apparent Image Quality (FLL-AIQ)
http://www.clarkvision.com/articles/...ex.html#FLLAIQ

And then there is the Focal Length Limited Apparent Image Quality Maximum (FLL-AIQ-MAX)
http://www.clarkvision.com/articles/...html#FLLAIQMAX
where the smaller pixels win. (e.g. see the moon image comparison I posted earlier in this thread: the 7D image is clearly superior to the 1D4 image).

But all this merely shows there are multiple ways to look at the problem. The original issue was driven by Arash's comment in this thread about the 7D being noisy in a focal length limited
situations:
http://www.birdphotographers.net/for...roated-Swallow

So that case, if the image were taken with a 1D4 and the same lens from the same position, the image would have less noise, but with less detail. Detail in this case was important and more reach was needed. That started this thread: a focal length limited situation.

Each camera is a tool, and each can produce better images in certain situations. In focal length limited situations where you need more reach, smaller pixels will win, pretty much always. And I still do not think the 7D should be bashed as a noisy camera.

I really don't think we need to discuss this further, and I'll bow out.

I will have a new article on Etendue and telephoto reach if you want more information and new insights night discussed here. Perhaps it can also be reproduced here on BPN. Maybe finished in a couple of weeks.

Roger

[Doug Brown]

I really don't think we need to discuss this further, and I'll bow out.

Me too!!!

[James Shadle]

Men,
You have got to dumb this down for me.

If I correctly understand what Roger is saying :

• An image from a 7D with a 500mm F4 lens will look close to an image from a Mark IV with a 500mm F4 lens if it has 1.4 converter attached.
• An image from a 7D with a 500mm F4 lens will look close to an image from a Mark IV with a 500mm F4 lens if the Mark IV image is cropped to match the 7D.
• An image from a 7D with a 500mm F4 lens will look close to an image from a Mark IV with a 500mm F4 lens if the ISO of the Mark IV is double the 7D's
• If your using a Mark IV with a 500mm F4 lens and you are able to have your subject the same size in the frame as a 7D with a 500mm F4, the Mark IV will have less noise.

The sunny 16 rule - Regardless of focal length, ISO=Shutter Speed @ F16 or any reciprocal value of that value. Right?

This is a 7D v Mark IV noise discussion. Not necessarily a 7D v Mark IV long lens noise test.
Photographing in low light @ ISO 1600 with a 24mm F2.8 lens, which body do you use?

[James Shadle]

Looks like I'm late to the party, everyone is bowing out.
Always a day late and a dollar short.