Technical Discussion: Resolution & Image Brightness

I've been following the discussion in the 5Ds/5Dsr announcement thread because this new camera will have a similar pixel density to many of the 18 MP crop cameras that were introduced 5 years ago (At the time I was very impressed with the level of detail that could be captured with a sensor of such pixel density). In the thread there has been a discussion about evaluating noise, and it got me thinking about the basics of making a digital recording of a scene with two sensors of same physical size but with two different densities. For example, a nominal 25 MP full-frame sensor and a nominal 50 MP full-frame sensor.

Let's assume for now that the photosites and imaging systems have the same properties, sensitivies, well depths, etc. Since the 50 MP sensor has twice as many pixels/photosites as the 25 MP sensor in the same unit physical area, if the sensors would presented the same exposure, the pixels of the 50 MP sensor would each collect half as many photons as the pixels of the 25 MP sensor. But if we compared the images, say on the rear LCD of same size, pixel density, brightness setting, etc., the images would appear the same brightness. Why? Shouldn't the 50MP image be darker because all signals were halved from collecting only half the number of photons? Is the 50 MP imaging system compensating with the application of a gain? Does the demosaicing technique involve summing nearby pixel values to arrive at the same brightness in that area? Or is it a different reason altogether?

I haven't heard of anyone pointing out that their images were getting darker because of moving to cameras with greater resolution. I'd like to understand what's going on here from forum residents that are more expert than I on this or if I've missed something in my thinking.

I haven't looked at the code which generates the preview images, but I'm certain that they normalize the brightness based on full well capacity of the sensor - or some value which amounts to the same thing.

This is done in hardware.

What comes off each sensel is not a count but a voltage, which has to be converted to a number via an analogue–digital converter (ADC). This is an electronic circuit which "knows" what the maximum voltage should be, taking into account the well size and analogue amplification (ISO) applied before. This produces a 14-bit number. That number ends up in the raw file as a (misnamed) "pixel" value.

From there software performs lens correction, demosaicing and conversion to a gamma space like sRGB or Prophoto RGB.

If you want a REALLY detailed technical discussion, check out this site:

http://clarkvision.com​/articles/index.html

Specifically, go here for discussion on sensor performance theory. There is a section talking about pixel density.

The guy has great credentials (MIT Ph. D. and NASA imaging scientist).

His whole site is a great read, especially if you are into low light and astrophotography.

Poe wrote in post #17446067
I've been following the discussion in the 5Ds/5Dsr announcement thread because this new camera will have a similar pixel density to many of the 18 MP crop cameras that were introduced 5 years ago (At the time I was very impressed with the level of detail that could be captured with a sensor of such pixel density). In the thread there has been a discussion about evaluating noise, and it got me thinking about the basics of making a digital recording of a scene with two sensors of same physical size but with two different densities. For example, a nominal 25 MP full-frame sensor and a nominal 50 MP full-frame sensor.

Let's assume for now that the photosites and imaging systems have the same properties, sensitivies, well depths, etc. Since the 50 MP sensor has twice as many pixels/photosites as the 25 MP sensor in the same unit physical area, if the sensors would presented the same exposure, the pixels of the 50 MP sensor would each collect half as many photons as the pixels of the 25 MP sensor. But if we compared the images, say on the rear LCD of same size, pixel density, brightness setting, etc., the images would appear the same brightness. Why? Shouldn't the 50MP image be darker because all signals were halved from collecting only half the number of photons? Is the 50 MP imaging system compensating with the application of a gain? Does the demosaicing technique involve summing nearby pixel values to arrive at the same brightness in that area? Or is it a different reason altogether?

I haven't heard of anyone pointing out that their images were getting darker because of moving to cameras with greater resolution. I'd like to understand what's going on here from forum residents that are more expert than I on this or if I've missed something in my thinking.

smaller pixels take less to fill up, just as a cup takes less to fill than a bucket. You can fill your yard with a few hundred buckets, or you can fill it with a few thousand carafes; after a rain the level in each will be the same. You can think of the "level" as the voltage output of each sensel, which is what correlates to the pixels brightness.

The scenario I've put forward isn't what you describe. Using your bucket analogy, it would be as if I set out 25 buckets in a given area then set out 50 buckets on the same area, with each group collecting same amount of rain in total but evenly distributed amongst the buckets in the group.

Poe wrote in post #17448282
The scenario I've put forward isn't what you describe. Using your bucket analogy, it would be as if I set out 25 buckets in a given area then set out 50 buckets on the same area, with each group collecting same amount of rain in total but evenly distributed amongst the buckets in the group.

Think of a large video display made up of lights (like at some sports arenas). You can have one high wattage lightbulb at each "pixel" location, or 4 lower wattage bulbs in the place of each large one. The total brightess is the same. Or for the rainfall analogy, one large bucket (one large pixel) catches the same amount of rain as 4 small buckets (4 small pixels) in the same area.

I don't disagree with you there. And when those 0.25x wattages are digitized through the ADC, i would expect them to be a darker tones than the pixel with 1x wattage. So a print or image on screen should be darker too when utilizing the more dense sensor.

Brightness is measured as a rate, e.g., in photons per sq mm per second. In the rainstorm, a small bucket collects less water, but if you weigh the water (like counting the photons or electrons etc in the well, because we don't know how "deep" they are) you need to know the collection area of the bucket to know the rainfall rate (brightness).

Poe wrote in post #17450019
I don't disagree with you there. And when those 0.25x wattages are digitized through the ADC, i would expect them to be a darker tones than the pixel with 1x wattage.

No. The ADC is measuring the voltage output from the amplifier (whose gain is set by the ISO setting), and the maximum voltage of that amplifier is whatever the camera designer felt like making it. (For example, this is probably higher on a Canon camera than a Sony camera, because Sony do everything on the sensor chip but Canon run that voltage signal along a circuit board and they will want it to be robust against noise.)

An ADC requires a reference voltage which corresponds to all 1s on its output. That voltage will be higher if the maximum output voltage from a pixel after amplification is higher. Whether it's higher because the camera has fewer pixels or because of the camera design doesn't matter.

AJSJones wrote in post #17450097
Brightness is measured as a rate, e.g., in photons per sq mm per second. In the rainstorm, a small bucket collects less water, but if you weigh the water (like counting the photons or electrons etc in the well, because we don't know how "deep" they are) you need to know the collection area of the bucket to know the rainfall rate (brightness).

I don't follow. What does the value of the luminous flux have to do with the imaging system determining the resulting brightness? I thought total charge accumulation in that pixel on the sensor determines brightness.

melcat wrote in post #17450391
No. The ADC is measuring the voltage output from the amplifier (whose gain is set by the ISO setting), and the maximum voltage of that amplifier is whatever the camera designer felt like making it. (For example, this is probably higher on a Canon camera than a Sony camera, because Sony do everything on the sensor chip but Canon run that voltage signal along a circuit board and they will want it to be robust against noise.)

An ADC requires a reference voltage which corresponds to all 1s on its output. That voltage will be higher if the maximum output voltage from a pixel after amplification is higher. Whether it's higher because the camera has fewer pixels or because of the camera design doesn't matter.

So you're saying the parameters of the amplifier and ADC are adjusted so that the accumulated charge in the pixel becomes the correct brightness value in the raw data and this will make the images from the sensors of two different pixel densities appear the same on the rear LCD, in lightroom, on a print, etc.?

Poe wrote in post #17452487
I don't follow. What does the value of the luminous flux have to do with the imaging system determining the resulting brightness? I thought total charge accumulation in that pixel on the sensor determines brightness.

Um, photon (i.e., luminous) flux IS the brightness that is recorded in an imaging system. Doesn't matter which way you go - collection or emission, it is photons PER AREA UNIT in a specified time not just the absolute number (photons/electrons/cha​rge etc) that determines brightness. Rainfall is measured in raindrops per square foot per second, not just raindrops per second.

Example. We have an image of a uniformly lit wall landing on an imaging device. It is sending 10,000 photons per square mm per second (made up numbers, of course). That is set (by the device maker - the answer to your question for melcat is YES - it is set, based on ISO "standards") to "brightness=100%" again arbitrary. We have one pixel that is 1 sq mm - it will pick up 10,000 photons in 1 sq mm in one second and the system says "Aha - that pixel recorded 100% brightness and the data is written as such." Now we can divide that pixel into 4 separate pixels. Each one will now get 2,500 photons and the system says "Aha, that pixel recorded 2500 in 1/4 of a sq mm so that's the same PHOTON FLUX as 10,000 per sq mm" and that is "100% brightness" so the data is written the same way. You wouldn't say that when we use a bucket of a quarter of the area that the rainfall rate (aka photon flux, aka brightness) is only 1/4 as much (even though we collected less rain) - or would you

How is it that can I achieve the same brightness with different fluxes? Shouldn't I only be able to achieve one brightness with one flux?

Poe wrote in post #17452659
How is it that can I achieve the same brightness with different fluxes? Shouldn't I only be able to achieve one brightness with one flux?

By changing the shutterspeed or aperture - those will affect how many of the photons in the incoming flux from the object will be captured or by changing the ISO, which alters how many captured photons will be set to 100% "brightness".

Thanks, AJSJones.