Why is sensor dynamic range defined as full-well capacity divided by read noise?

Asked 7/2/2014

15 views

2 answers

0

I understand that a sensor’s dynamic range is often described as the ratio of a pixel’s full-well capacity to its read noise. For example, if a photosite can hold about 44,900 electrons and the read noise is 10 electrons, the dynamic range is about 44,900/10 = 4,490:1.

What I don’t understand is why that ratio is sometimes interpreted as meaning there are 4,490 gray levels available. Why does dividing capacity by read noise describe dynamic range, and how is that different from the number of discrete tonal or grayscale levels a camera can record?

Originally by Photography Stack Exchange contributor. Source · Licensed CC BY-SA 4.0

Photography Stack Exchange contributor

12y ago

2 Answers

7

Simple: you're confusing the term range with the idea of number of levels. That's not the case. The value in the dynamic range ratio says nothing about the number of individual gray levels available within that range.

The range is the span, not the number of discrete bits. A ratio is used (in your example, 44900:10, or approximately 4500:1), because the actual number of photons counted is not the information we're concerned with here — we just want to know the relationship between the brightest and then darkest possible values. (Here, the brightest value is 4500× the darkest.)

This can also be expressed in stops, which is simply applying a base-2 logarithmic scale to this same ratio. That is, 4500:1 is approximately 12 stops (because 4500 is slightly greater than 2¹².) Or, to put it another way, with this amount of dynamic range, you can capture 12 doublings of the brightness.

Within each of those doublings, you could use a single value per band, resulting in a hard edge between each band of brightness. Or you could use tens, hundreds, or thousands of gray levels per band, for a smooth grayscale.

Or you could use one bit, flipped to either "bright" or "dark" — and the ratio between those two levels would still be the dynamic range.

Here's an example of that. The following image has only two different values, and let's say (picking a number that might be reasonable for an LCD screen) the dynamic range between the blackest black and the whitest white is 1:1000.

black and white

Imagine that the dark side is the lowest level that can be reasonably read, and the light side is the brightest.

Now, here's the same thing, but with 255 values:

black, gray, and white

The dynamic range here is the same — the ratio between that darkest dark and the whitest white. But, obviously, there a lot more discrete gray levels in between. Since this is an 8-bit PNG, it happens that there are 256 total. But that's an entirely separate thing.

Of course, the more dynamic range you have, the more bits are necessary to provide a nice, smooth gradient. But it's not mandated in any way, or part of the definition.

Originally by user1943. Source · Licensed CC BY-SA 4.0

user1943

12y ago

0

AI Answer

Generated from our catalog & community — verify before relying on it.

Because dynamic range is a ratio of signal limits, not a count of tonal steps.

In this example, the largest usable signal is the pixel’s full-well capacity (44,900 electrons). The smallest usable signal is limited by read noise (about 10 electrons). So the sensor can distinguish signals across a span of roughly 44,900 ÷ 10 = 4,490:1.

That does not mean the camera only has 4,490 gray levels. It means the brightest usable signal is about 4,490 times stronger than the darkest usable signal above the noise floor.

The number of gray levels is determined by digitization/bit depth and processing, not by the dynamic-range ratio itself. Dynamic range tells you how wide the brightness span is; bit depth tells you how finely that span can be divided.

This is also why dynamic range is often expressed in stops: taking log base 2 of the ratio. A 4,490:1 range is about 12 stops, meaning the brightest usable value is roughly 12 doublings above the darkest usable value.

UniqueBot

AI

12y ago

Your Answer