this chart is hard for me to make any sense of.
Maybe you would explain how to use it, or summarize the data therein
I decided to do that only in case someone asks for it. Well, you did. (Later I will expand the chart with explanation).
The noise depends on two factors:
1. the amount of captured and converted light, which becomes apparent for us in form of raw pixel values,
2. ISO gain,
The exposure duration too is a factor, but only with longer exposures (whatever "longer" means, it is not under one second; the sample images are always in fractional seconds).
Note: the noise does not depend on the color.
Noise is measured on smooth, unicolored, evenly lit spots as the standard deviation, which is an absolute value. That value can not be directly compared between cameras for several reasons; the simplest to think of is the bit depth. Thus the numerical range can vary widely, like {0-4096} or {0-16363}, or {128-3600} etc.
So, the raw (non-demosaiced) pixel values get linearly mapped on the uniform range of {0-255}; this step takes care of black levels and different saturation points; it can be seen as "normalization". The noise is then expressed as percentage of the standard deviation in the normalized intensity of the average pixel values of the selected "color channel" in the selected area. This is the
NR column.
The amount of captured light is expressed in the "relative darkness" of the selected spot. This is very important, for the same degree of noise is much worse on a well-lit area than in the very shadows. The relative darkness too needs to be normalized (again, the average pixel values can not be compared directly between cameras, in many cases not even between ISOs of the same camera). The measure of the intensity is "dynamic range position", i.e. how many stops it is from "white" (from pixel saturation). This is the
DR column.
The dynamic range of a certain camera at a certain ISO can be found by locating the noise level, which is deemed acceptable. Later I plan to create a "chart of noises" to illustrate, which noise level has what appearance (this is not equal to what one sees in Phoptoshop under the statistics on the Histogram panel). Here is an example for 19.9% noise; it happens to occur in an image from the 5D2 @ ISO 3200, in the 5.91st stop (in the green channel, but that plays no role):
5D2 noise sampleA camera is more noisy than another one (or the same one at another ISO) if the noise level is higher with comparable intensities, or the intensity is higher at comparable noise levels (the better camera reaches a certain noise level is on darker spots, right?).
The difference between two cameras or two ISOs can be quantified by looking for the same noise level.
Example: the Canon 40D shows noise level 15% at the 8th stop @ ISO 100 (the best ISO of that camera). The Nikon D3's best ISO is 200 (?); the 15% noise occurs at the 8.7th-8.9th stop, i.e. it beats the 40D by at least 3/4 stop.
Comparison between different ISO values of the same camera reveals the loss of DR incurred by the increased ISO gain. If the loss reaches one stop, then there is no point to use that ISO (except for JPEG shooters).
There are many contradicting value pairs in the chart, due to the less than ideal sample shots. For example Imaging Resources use color checker card and other charts, which look like a puppy had been playing with them and then the peed on them. (One does not see this in normal view.) The Nikon D3 values are the most reliable (measured on a Stouffer transparenvy wedge).
I don't see any MF backs on there. Am I missing something?
Yes, and so do I, namely raw images suitable for measurement.