[kernos]

I was talking to the wedding photographer for my son's wedding and looking at his gear - Canon 5d Mark IV and a Mark II as backup. He said that other folks on the team had taken the Canon 5DS R but he was just as glad to be working with the older technology because the R allowed only about a 1/3 of a stop of exposure latitude. The guy's work speaks for itself, so I'm not doubting him, but I wonder why more tightly packed pixels would affect exposure latitude. Any thoughts on this?

[enoeske]

I was talking to the wedding photographer for my son's wedding and looking at his gear - Canon 5d Mark IV and a Mark II as backup. He said that other folks on the team had taken the Canon 5DS R but he was just as glad to be working with the older technology because the R allowed only about a 1/3 of a stop of exposure latitude. The guy's work speaks for itself, so I'm not doubting him, but I wonder why more tightly packed pixels would affect exposure latitude. Any thoughts on this?

I think he's refering to the dynamic range of the sensors. The higher res sensor in the R has less dynamic range than the IV.

[kernos]

That would make sense.

[biz-engineer]

Digital sensors are all mostly ISO invariant, meaning that the latitude peaks around 150-300 ISO, and then drops as the ISO setting gets higher than 400. Over the years, I came to the conclusion that the key to good image result came down to keep ISO rather low within a limited range, and that the resolution of the sensor was secondary. Keeping ISO within a low range is achieved by using fast enough lenses, flash strobes or the use of a good tripod, given that lighting is fixed by external conditions that are independent from the camera system we are using. For photographing moving subjects, such as for wedding, tripod or SR aren't valid options, use of faster lenses or strobes help a lot to keep ISO within a low range. The actual difference of sensor latitude between various sensors is something like 0.3ev to 0.5ev, that is not much compared to being able to have 2 stop of extra latitude by using a lens with f2.8 max aperture instead of a lens with f5.6 aperture. The same is true when shooting with a f1.4 lens vs f2.8, two stops of exposure difference returns a more significant step up in image quality compared to the small difference between two camera models.

---

Last edited by biz-engineer; 11-06-2018 at 11:25 AM.

[kernos]

This photographer, Adam Bird of Chicago, had a good chunk of the venue (wedding and reception together) covered with strobes, so he was able to have his cake and eat it too.

[photoptimist]

This says more about Canon's sensors than it does about any hard-baked relationship between resolution and exposure latitude (which is the span of forgiveness to over-exposing or under-exposing the image).

Most of the modern sensors made by Sony are iso-invariant and have a very wide dynamic range. That means that if the base ISO is 100 and ISO 1600 images are usable, the exposure latitude at ISO 100 is at least 4 stops. It's actually a little higher than that because the 4 stops is just the under-exposure side of the latitude. It is true that digital sensor tend to have poor over-exposure latitude compared to print film but the total latitude is quite large.

[ProfessorBuzz]

One of the issues at play is that there's a tradeoff between sensor size and pixel size - more resolution = more pixels in the same physical sensor dimensions, so that means the pixels are smaller, have less area, and thus less light-gathering capability.

[biz-engineer]

One of the issues at play is that there's a tradeoff between sensor size and pixel size - more resolution = more pixels in the same physical sensor dimensions, so that means the pixels are smaller, have less area, and thus less light-gathering capability.

For a given sensor size more pixels are smaller, but they are more pixels hopefully collaborate together, they kinda "help each other" in making the final image :-)

[Bob 256]

When smaller pixels are used, less light is received by each pixel. This can be compensated for by amplifying the output signal more but that's about the same thing that's done in raising ISO for a given sensor, so dynamic range suffers as mentioned previously. You also raise the noise levels. Think of a sponge that's 1 inch think. It has a certain grain to it. If that sponge is cut in half but is stretched to fill the 1 inch thickness as before, the "grain" increases. The amount of noise limits what can be brought out of the shadow areas so less shadow information can be retrieved when the sensor is starved for light and the signal is amplified to compensate. Smaller pixels - less light per pixel (for a given image size).

[photoptimist]

When smaller pixels are used, less light is received by each pixel. This can be compensated for by amplifying the output signal more but that's about the same thing that's done in raising ISO for a given sensor, so dynamic range suffers as mentioned previously. You also raise the noise levels. Think of a sponge that's 1 inch think. It has a certain grain to it. If that sponge is cut in half but is stretched to fill the 1 inch thickness as before, the "grain" increases. The amount of noise limits what can be brought out of the shadow areas so less shadow information can be retrieved when the sensor is starved for light and the signal is amplified to compensate. Smaller pixels - less light per pixel (for a given image size).

It may be true that smaller pixels are somewhat grainer but those more intense "grains" are smaller and thus less noticeable. Those two effects largely cancel each other. Thus, the dynamic range (and latitude) of the pixels might be getting worse by the dynamic range (and latitude) of the image is staying much the same.

You can see how this works by considering how B&W film works. With film, the individual "pixels" are tiny silver-halide grains. The dynamic range of an individual grain totally sucks. No matter what the light level is, an individual developed grain will either develop fully black or fully white. It's only the probability of it being black or white that changes. A film grain is the equivalent of a silicon pixel so small it saturated after only 1 electron. But make these low-DR grains small and put lots of them together and all those pure black speckles visually merge to make a smooth gray with good dynamic range at the image level.

Thus, the smaller size of the high-resolution pixels compensates for the noisiness of the pixel through numbers.

[biz-engineer]

When smaller pixels are used, less light is received by each pixel. This can be compensated for by amplifying the output signal more but that's about the same thing that's done in raising ISO for a given sensor, so dynamic range suffers as mentioned previously

How about the Pentax K1 having a smaller pixels and better dynamic range than the Canon 5D mark III ?

[Rondec]

When smaller pixels are used, less light is received by each pixel. This can be compensated for by amplifying the output signal more but that's about the same thing that's done in raising ISO for a given sensor, so dynamic range suffers as mentioned previously. You also raise the noise levels. Think of a sponge that's 1 inch think. It has a certain grain to it. If that sponge is cut in half but is stretched to fill the 1 inch thickness as before, the "grain" increases. The amount of noise limits what can be brought out of the shadow areas so less shadow information can be retrieved when the sensor is starved for light and the signal is amplified to compensate. Smaller pixels - less light per pixel (for a given image size).

I don't think size of pixels is a big factor. The only thing that smaller pixels does is drives pixel peepers crazy. Because you can zoom in more on your image, you see more flaws and noise. But the reality is the only way to compare images is to pick a display size and then to compare them at that size. Once you do that, the smaller pixels get binned together and the noise advantage of the larger pixels goes away.

The bigger problem that the OP is talking about has to do with Canon sensors. If you look at DXO Mark scores and graphs, you can see that Sony sensors really do well regardless of pixel density and a sensor like the D850's 45 megapixel one performs better with regard to dynamic range than the Sony A9's 24 megapixel sensor does.

[Bob 256]

How about the Pentax K1 having a smaller pixels and better dynamic range than the Canon 5D mark III ?

I probably should have qualified my example. I was talking about effects over a single sensor technology. Different technologies can have improvements (like making a sponge with finer texture) resulting in sensors with less noise than others. That leads to the ability to throw more amplification at a sensor and not get as much noise (or dynamic range reduction) as a comparative sensor. Just look at sensors from 10 years back and compare those to the K-1 sensor we have today. However, if we talk about just one sensor technology, generally speaking, denser and smaller (not necessarily more) pixel sites lead to poorer sensor performance.

If the exact same pixel technology as exists in the K-1 were used to create a 144Mpx sensor by shrinking the pixel sites (same sensor size overall), one would expect worse performance (higher resolution however). Crop that sensor to get back to the 36 Mpx K-1 resolution, and the decrease in performance would remain (with a sensor which is physically smaller). (This all assumes the 36 Mpx pixel architecture could be shrunk by a factor of two to get 144Mpx).

[photoptimist]

If the exact same pixel technology as exists in the K-1 were used to create a 144Mpx sensor by shrinking the pixel sites (same sensor size overall), one would expect worse performance (higher resolution however). Crop that sensor to get back to the 36 Mpx K-1 resolution, and the decrease in performance would remain (with a sensor which is physically smaller). (This all assumes the 36 Mpx pixel architecture could be shrunk by a factor of two to get 144Mpx).

But that's not true. Although the individual pixels on the 144 Mpix sensor might be noisier, that noise is cancelled out by the smaller size and greater numbers of pixels.

Display the 36 MPix image side-by-side with the 144 MPix image on a 4k display or make a 36" x 24" prints and the noise levels and dynamic range will be almost identical in both images. Although the 144 MPix image has noisier pixels, four times as many sensor pixels are averaged together for each display pixel of the 4k display or each visual spot on the print.

If anything, the 144 MPix image will have higher quality because it will resolve more detail.

[Bob 256]

But that's not true. Although the individual pixels on the 144 Mpix sensor might be noisier, that noise is cancelled out by the smaller size and greater numbers of pixels.

Display the 36 MPix image side-by-side with the 144 MPix image on a 4k display or make a 36" x 24" prints and the noise levels and dynamic range will be almost identical in both images. Although the 144 MPix image has noisier pixels, four times as many sensor pixels are averaged together for each display pixel of the 4k display or each visual spot on the print.

If anything, the 144 MPix image will have higher quality because it will resolve more detail.

I don't disagree with what you're saying. More pixels will contribute to the "dither" of the original single pixel space since the noise element in each is random, however the randomness from the actual value that pixel space should have is larger (higher signal noise) for each of the smaller pixels. I'd have to open my statistics book to see how these would combine, but I believe you are correct in saying that four pixels added (actually averaged) with more noise (and that noise won't be twice the single pixel noise) will contribute to a better image, both noise and resolution wise.

There does come a point however, where reduction in the size of the pixels so much starves each one for light, that performance will fall off considerably. I have no idea if modern cameras are at that point yet (apparently not since the K-1 is such a great performer).