fredmiranda.com Login
This topic is in the 2015 archive. You can read it and like its posts here. It is read only.

Archive 2015No DR improvement = no 5Ds for me

  
 
alundeb
Offline
Details
p.10 #1No DR improvement = no 5Ds for me
RustyBug wrote:
alundeb wroteBut takosawa also brings in read noise here, which is important for (engineering) dynamic range.
Go on ...


I am an en engineer with scientific training and a specialist in signal processing, but I am not a good pedagogue. I don't think I can explain this better that what these sources do. I have written about it many many times on this and the alternative board, though, but I don't always get through to everybody.
Feb 08, 2015 at 01:29 PM
takowasa
Offline
Details
p.10 #2No DR improvement = no 5Ds for me
alundeb wrote:
But takosawa also brings in read noise here, which is important for (engineering) dynamic range.


RustyBug wrote:
Go on ... "read noise"


Read noise is the additional electronic noise from the sensor and other electronics in the camera. It is generally very small compared to the photon noise (the noise from the light itself) and thus only becomes a factor for portions of the photo made with very little light (dark shadows and high ISOs).

Feb 08, 2015 at 01:30 PM
takowasa
Offline
Details
p.10 #3No DR improvement = no 5Ds for me
RustyBug wrote:
ISO 1600 = amplification

Sorry, I don't have a better technical explanation to offer (maybe others can assist).


takowasa wrote:
So you're saying that the "amplification" of ISO 1600 causes the noise, not the fact that 1/16 as much light is falling on the sensor? What about 1/1600 ISO 100, then? No amplification there.


RustyBug wrote:
I'm assuming you have retained the same shutter speed to result in a corresponding 4 stop reduction in light.


No, 1/16 the shutter speed. Again, same camera, same scene, one photo at 1/100 ISO 100, the other at 1/1600 ISO 1600 or 1/1600 ISO 100 (then pushed four stops to achieve the "correct" brightness).

Off the cuff ... 4 stops underexposure could be lacking detail @ near silhouette for portions of the scene (depending on DR of the scene), but it could still be clean without any amplification to increase the noise (layman perspective only).

The 1/1600 photos (either ISO 1600 or ISO 100) are both underexposed four stops compared to the 1/100 photo. That means that they are made with 1/16 as much light since we're using the same sensor for all the photos. Thus, they will be more noisy.

So, once again, the noise in a photo is determined entirely by how much light makes up the photo, all else equal. If we make the pixel counts unequal, then, at the pixel level, the photo with more pixels will be more noisy, but not at the image level, as the photos in the link I posted before demonstrate:

http://www.dpreview.com/forums/post/37549285

Of course, there is the consideration that more smaller pixels tend to have greater electronic noise than fewer larger pixels, and thus a more noisy photo overall for the portions of the photo made from very little light (e.g. deep shadows at base ISO or high ISO).

On the other hand, as the photos in the link above also show, the application of noise filtering can serve to tip the balance in favor of the more numerous smaller pixels, since the photo made with more smaller pixels will be more detailed and give the noise filtering more to work with.
Feb 08, 2015 at 01:39 PM
jcolwell
Offline
Details
p.10 #4No DR improvement = no 5Ds for me
jcolwell wrote:
takowasa, I really think you're on the wrong track.

takowasa wrote:
Why, in your opinion, then, does a photo of a scene with the same camera, at 1/100 ISO 100 have less noise than a photo at 1/1600 ISO 1600 (or 1/1600 ISO 100 pushed 4 stops), if it is for a reason other than the fact that 16x as much light made up the photo with an exposure time of 1/100?


Are you saying that a photo taken with t = 1/100, f/4, and ISO 100 will have different noise characteristics than the same photo taken with t = 1/50, f/8, and ISO 100 ?

The reason that ISO 1600 images have more noise is because they have a much higher gain applied to the output from the sensor.

In my opinion, there are two primary reasons for noise diffrences between FF and crop sensors that have the same pixel density, i.e. pixels per mm (ppmm):

(i) the FF sensor is usually in a camera that has a generation newer tech than the crop factor camera - for example, the 20D (2004) and 1DsIII (2007) have the same pixel density. In the current case, where we're discussing the 5Ds/5DsR and 7DII, this might not be relevant, as there's only about eight months between the 7DII release date and the announced release date for the 5Ds. Time will tell.

(ii) at a common viewing size (with images having the same "framing", and so using different focal length lenses), the FF sensor image is enlarged significantly less than the crop factor image (especially 1.6x CF), and so noise is a relatively smaller part of the image (which is opposite to something you said earlier). For example, with the 5Ds sensor at 8688 px wide and the 7DII sensor at 5472 px wide, a print 40" wide (at 300 dpi) has a magnification of 38% for the 5Ds and 119% for the 7DII. Hence, the noise is much more pronounced.

YMMV
Feb 08, 2015 at 02:16 PM
takowasa
Offline
Details
p.10 #5No DR improvement = no 5Ds for me
takowasa wrote:
Why, in your opinion, then, does a photo of a scene with the same camera, at 1/100 ISO 100 have less noise than a photo at 1/1600 ISO 1600 (or 1/1600 ISO 100 pushed 4 stops), if it is for a reason other than the fact that 16x as much light made up the photo with an exposure time of 1/100?


jcolwell wrote:
Are you saying that a photo taken with t = 1/100, f/4, and ISO 100 will have different noise characteristics than the same photo taken with t = 1/50, f/8, and ISO 100 ?


No. I'm talking about using the same camera to take a photo of the same scene with the same lens and aperture. So, 1/100 ISO 100 "properly exposed" with one camera, 1/1600 ISO 1600 (or 1/1600 ISO 100 pushed to match the "correct" brightness), with the same scene, camera, and aperture.

Why does the 1/100 photo have less noise?
Feb 08, 2015 at 02:21 PM
jcolwell
Offline
Details
p.10 #6No DR improvement = no 5Ds for me
...because the output from the sensor has a gain of unity at ISO 100 (for a base 100 ISO camera), and the output from the sensor has a gain of 4 EV (2^4 = 16) at ISO 1600, and so there's a lot more noise.
Feb 08, 2015 at 02:25 PM
alundeb
Offline
Details
p.10 #7No DR improvement = no 5Ds for me
jcolwell wrote:
The reason that ISO 1600 images have more noise is because they have a much higher gain applied to the output from the sensor.

YMMV


Sorry, no.

Say at ISO 100, at average 10000 photons are converted to electrical charge in each pixel. The photon shot noise is then sqrt(10000) = 100 photons, and the signal/noise ratio is 10000/100 = 100

Then at ISO 1600, at average 10000/16 = 625 photons are converted. The photon shot noise is then sqrt(625) = 25 photons, and the signal/noise ratio is 625/25 = 25.

The image at ISO 100 has 4 times better signal to noise ratio than the image at ISO 1600 because of the amount of light, and amplification is not in the equation.
Feb 08, 2015 at 02:26 PM
jcolwell
Offline
Details
p.10 #8No DR improvement = no 5Ds for me
OK.

I still think my points about relative sensor tech and (especially) image magnification are correct.
Feb 08, 2015 at 02:27 PM
jcolwell
Offline
Details
p.10 #9No DR improvement = no 5Ds for me
Hang on a second. I was replying specifically to this post,

https://www.fredmiranda.com/forum/topic/1346234/8#12847964

I was assuming that the same illumination was present for both images, e.g. f/2.8 for ISO 100 and f/11 for ISO 1600. In which case both have the same amount of light. How many things can you can change between comparison cases and still make sense of what's really going on?

That's why my next post said, ...

"Are you saying that a photo taken with t = 1/100, f/4, and ISO 100 will have different noise characteristics than the same photo taken with t = 1/50, f/8, and ISO 100 ?"

...same amount of light.

The "same aperture" provision wasn't mentioned (at least not that I saw) until after I made that "gain" post.

Complex situation.
Feb 08, 2015 at 02:38 PM
snapsy
Offline
Details
p.10 #10No DR improvement = no 5Ds for me
There are two main types of noise in digital images, photon shot noise and read noise. Shot noise is the predominate noise at most exposures and read noise predominates in the shadows. As ISO increases the tones that were previously in the shadows at lower ISOs start sliding into the midtones and thus read noise progressively has a larger effect on the total noise as ISO is increased. Regarding ISO and its relationship to noise, it's not the higher ISO itself that causes additional noise but instead the lower absolute exposure (aperture + shutter speed), which results in less light reaching the sensor. Increasing the nominal ISO is applying gain to the signal (analog and/or digital) in order to produce an acceptable output brightness. In other words, ISO is not part of the actual exposure in technical terms and has little effect on total noise performance outside of the analog vs digital gain differences present on some legacy sensor designs such as those from Canon.

Regarding pixel size/density and its correlation to the two types of noise described (shot noise and read noise), pixel size generally has no effect on shot noise but does affect read noise. Since shot noise predominates most exposures/ISOs, this means pixel size does not have an effect on most images. You can see evidence of this by comparing sensors of the same size and efficiency but with differing pixel counts. For example, the 36MP D800 has almost identical SNR (primarily a shot noise measurement) to the 12MP D3s because even though the D800 has 3x the pixels it has the same underlying quantum efficiency, thus both sensors convert the same amount of light into electrons even though their pixel counts are wildly different. This is based entirely on image-level measurements where the same portion of the sensor is compared irrespective of the number of pixels. This is why the High ISO performance of the D800 is nearly identical to the D3s in terms of midtone noise, where shot noise predominates.

Regarding read noise, there is a dichotomy between DR at base ISO and High ISO due to the differences in conversion gain required between a large pixel, high full-well-capacity (FWC) sensor like the D3s/1DX/Df vs higher density sensors like the Sony 24MP/36MP Exmors. In a large pixel sensor, the FWC (how much light/electrons each individual pixel can hold) is much higher, resulting in a higher amount of referred read noise, which translates to lower base ISO DR, but this disadvantage turns into an advantage at Higher ISOs due to both the higher conversion gain of the sensor from its larger FWC and also the lower total read noise aggregate from fewer pixels that must be read. The opposite is the case in smaller pixel sensors, as they have shallower FWC resulting from their smaller pixels. This again is demonstrated in the D800 vs D3s; at base ISO the D3s has 12EV of DR vs the D800's 14.3EV but at ISO 25,600 the D3s has 8EV of DR vs the D800's 7EV.

Last point - the general relationship of pixel count to read noise is a function of current technology and is subject to change. For example, the 16MP sensor introduced in the D4 employs a dual-gain method that allows it to select a low conversion gain at base ISO and a high conversion gain above that, offsetting some of the penalty of the larger FWC of its larger pixels (Aptina DR-Pixel Whitepaper (PDF)).

Edited on Feb 08, 2015 at 07:34 PM · View previous versions
Feb 08, 2015 at 02:46 PM
takowasa
Offline
Details
p.10 #11No DR improvement = no 5Ds for me
takowasa wrote:
I'm talking about using the same camera to take a photo of the same scene with the same lens and aperture. So, 1/100 ISO 100 "properly exposed" with one camera, 1/1600 ISO 1600 (or 1/1600 ISO 100 pushed to match the "correct" brightness), with the same scene, camera, and aperture.

Why does the 1/100 photo have less noise?


jcolwell wrote:
...because the output from the sensor has a gain of unity at ISO 100 (for a base 100 ISO camera), and the output from the sensor has a gain of 4 EV (2^4 = 16) at ISO 1600, and so there's a lot more noise.


Do you know why the "sensor gain" results in more noise? Because it amplifies the photon noise that results from the lesser amount of light falling on the sensor to achieve the brightness of the 1/100 photo. So, once again, it is the lesser amount of light falling on the sensor that results in the greater noise.

For example, let's say 3200 photons fell on the pixel for the 1/100 ISO 100 photo, and the sensor had a quantum efficiency of 50%. Then the signal would be 1600 electrons, and the photon noise would be sqrt(1600) = 40 electrons, resulting in a photon noise of 40/1600 = 2.5%.

For the 1/1600 photo (regardless of the ISO), the signal would be 100 electrons (1/16 as much) and the photon noise would be sqrt(100) = 10 electrons, making for a photon noise of 10/100 = 10% -- four times as much.

This disregards the effect of electronic noise, but electronic noise is on the order of 2 electrons at high ISOs, and and as high as 30 electrons at base ISO for Canon cameras (the high electronic noise at base ISO for Canon sensors is why their DR is so much less than the competition).

So, for portions of the photo made with very little light, we would have to include the electronic noise. But even for a signal as weak as 100 electrons, it wouldn't make much of a difference at ISO 1600 (although it would make a significant difference for ISO 100 pushed four stops with a Canon sensor).
Feb 08, 2015 at 02:47 PM
takowasa
Offline
Details
p.10 #12No DR improvement = no 5Ds for me
snapsy wrote:
There are two main types of noise in digital images, photon shot noise and read noise. Shot noise is the predominate noise at most exposures and read noise predominates in the shadows. As ISO increases the tones that were previously in the shadows at lower ISOs start sliding into the midtones and thus read noise progressively has a larger effect on the total noise as ISO is increased. Regarding ISO and its relationship to noise, it's not the higher ISO itself that causes additional noise but instead the lower absolute exposure (aperture + shutter speed), which results in less light
...Show more →

Nicely stated.
Feb 08, 2015 at 02:47 PM
takowasa
Offline
Details
p.10 #13No DR improvement = no 5Ds for me
jcolwell wrote:
Hang on a second. I was replying specifically to this post,

https://www.fredmiranda.com/forum/topic/1346234/8#12847964

I was assuming that the same illumination was present for both images, e.g. f/2.8 for ISO 100 and f/11 for ISO 1600.


I'm not sure why you would have thought such a thing. From the link you gave above:

Why, in your opinion, does a photo of a scene with the same camera, at 1/100 ISO 100 have less noise than a photo at 1/1600 ISO 1600 (or 1/1600 ISO 100 pushed 4 stops), if it is for a reason other than the fact that 16x as much light made up the photo with an exposure time of 1/100?

In any event, I hope things are clear now. I am talking about the same scene, the same camera, the same lens, and the same aperture.
Feb 08, 2015 at 02:51 PM
jcolwell
Offline
Details
p.10 #14No DR improvement = no 5Ds for me
takowasa wrote:
Do you know why the "sensor gain" results in more noise? ...


I know better now than I did before snapsy's post. OTOH, it looks like the situation I was addressing isn't the same situation as you were describing. Go figure.

Feb 08, 2015 at 02:53 PM
jcolwell
Offline
Details
p.10 #15No DR improvement = no 5Ds for me
takowasa wrote:
...In any event, I hope things are clear now. I am talking about the same scene, the same camera, the same lens, and the same aperture.


Yes, they are clear now. Unfortunately, you didn't specify the "same aperture" condition before I made those posts. My bad if you did mention it in a previous post.
Feb 08, 2015 at 02:56 PM
borno
Offline
Details
p.10 #16No DR improvement = no 5Ds for me
snapsy wrote:
There are two main types of noise in digital images, photon shot noise and read noise. Shot noise is the predominate noise at most exposures and read noise predominates in the shadows. As ISO increases the tones that were previously in the shadows at lower ISOs start sliding into the midtones and thus read noise progressively has a larger effect on the total noise as ISO is increased. Regarding ISO and its relationship to noise, it's not the higher ISO itself that causes additional noise but instead the lower absolute exposure ..........[/url]).


You've got a gift for putting thing in terms even I can get, snapsy.
Feb 08, 2015 at 04:07 PM
RustyBug
Offline
Details
p.10 #17No DR improvement = no 5Ds for me
takowasa wrote:
I don't know what to tell you. You asked for an example of the same pixel density on FF and crop, I gave one, and FF came up with less noise. You dismiss the example, feeling that there are "other reasons besides pixel density contributing to the noise difference" and ask for another example, with cameras that have "consumer level camera build".


Please understand, I'm not totally dismissing them, I can see the diff ... but as an honest question, iirc that is semi-pro vs. consumer entry grade components. In another dialogue elsewhere (can't recall where atm) an argument could be in reference to the impact it could effect noise depending on quality of components, heat buildup, power, etc.

So, aside from the 6D, what FF camera has a "consumer level build" that will satisfy you? Unfortunately, there are no 8 MP crop camera with sensors from the same generation that I can compare the 6D to.

Okay, that's what I was asking @ other examples. Just wondering if one existed, preferably @ semi-pro vs. semi-pro or pro vs. pro ... or even semi-pro vs. pro would seem preferable to including consumer grade build. Lots of ways to cut corners there.

Well, we do have the 5DS now, yes? When we compare it to the 7D2, we'll find the exact same situation as the D800 vs the D7000.

Well, at least then it will be two of the same series build. Just wondering if would get a bit closer on the apples being compared, not entirely refuting the applesauce.

Feb 08, 2015 at 04:25 PM
RustyBug
Offline
Details
p.10 #18No DR improvement = no 5Ds for me
alundeb wrote:
jcolwell wroteThe reason that ISO 1600 images have more noise is because they have a much higher gain applied to the output from the sensor.

YMMV
Sorry, no.

Say at ISO 100, at average 10000 photons are converted to electrical charge in each pixel. The photon shot noise is then sqrt(10000) = 100 photons, and the signal/noise ratio is 10000/100 = 100

Then at ISO 1600, at average 10000/16 = 625 photons are converted. The photon shot noise is then sqrt(625) = 25 photons, and the signal/noise ratio is 625/25 = 25.

The image at ISO 100 has 4 times better signal to noise ratio than the image at ISO 1600 because of the amount of light, and amplification is not in the equation.


Okay, so I'm following the sqrt numbers.

But, how do you get from 625 photons of energy to generate a signal equal to the signal generated by 10,000 photons of energy ... without some form of amplification or gain in the process?



Edited on Feb 08, 2015 at 06:12 PM · View previous versions
Feb 08, 2015 at 06:11 PM
takowasa
Offline
Details
p.10 #19No DR improvement = no 5Ds for me
takowasa wrote:
I don't know what to tell you. You asked for an example of the same pixel density on FF and crop, I gave one, and FF came up with less noise. You dismiss the example, feeling that there are "other reasons besides pixel density contributing to the noise difference" and ask for another example, with cameras that have "consumer level camera build".


RustyBug wrote:
Please understand, I'm not totally dismissing them, I can see the diff ... but as an honest question, iirc that is semi-pro vs. consumer entry grade components. In another dialogue elsewhere (can't recall where atm) an argument could be in reference to the impact it could effect noise depending on quality of components, heat buildup, power, etc.


The two primary attributes of a sensor that contribute to noise are its quantum efficiency (the probability that a photon falling on the sensor releases and electron) and the electronic noise added by the sensor and other hardware in the imaging pipeline.

For a given generation of sensors, the quantum efficiency is pretty much the same across the board, but the electronic noise can vary quite a bit, not only across brands, but as a function of the ISO setting. At high ISOs, however, there variation between sensors with regards to electronic noise diminishes quite a bit (again, for a given generation of sensors).

There are other factors involved besides quantum efficiency and electronic noise, but those are the two primary players in most situations.

takowasa wrote:
So, aside from the 6D, what FF camera has a "consumer level build" that will satisfy you? Unfortunately, there are no 8 MP crop camera with sensors from the same generation that I can compare the 6D to.


RustyBug wrote:
Okay, that's what I was asking @ other examples. Just wondering if one existed, preferably @ semi-pro vs. semi-pro or pro vs. pro ... or even semi-pro vs. pro would seem preferable to including consumer grade build. Lots of ways to cut corners there.


There is no case that I am aware of where a high end camera has a sensor with better tech than a low end camera. If anything, it's often just the opposite (cell phone sensors have the latest and greatest tech, for example).

takowasa wrote:
Well, we do have the 5DS now, yes? When we compare it to the 7D2, we'll find the exact same situation as the D800 vs the D7000.


RustyBug wrote:
Well, at least then it will be two of the same series build. Just wondering if would get a bit closer on the apples being compared, not entirely refuting the applesauce.


They will be close, but there may certainly be differences. For example, Canon did a lot to ameliorate the banding in the 6D sensor compared to the 5D3 sensor, even though the sensors are otherwise pretty much the same. Presumably, this is because the 6D sensor came out a tad later than the 5D3 sensor. So, it's not a matter of high end vs low end cameras, but when the sensors were designed.
Feb 08, 2015 at 06:12 PM
RustyBug
Offline
Details
p.10 #20No DR improvement = no 5Ds for me
5D, 6D & 7D are (imo) same grade of build ... just different feature sets. Different from XXD or XXXD series.
Feb 08, 2015 at 06:16 PM
FM ForumsCanon ForumJoin Upload & Sell
  
 

You are not logged inLog inRegister