Monday, January 24, 2011

Light Density – Does it make a difference to noise?

Well, by golly, it does. First off, what is light density? Well, it is the number of photons hitting the sensor. More photons hitting the sensor in any given interval of time means denser light. You can think of this as the “flow” of photons in much the same way as a pipe carrying more water or a wire carrying more electrons (current.)

The exposure, then, is determined by the shutter speed and the aperture, which is really saying the number of photons that hit the sensor in the time the shutter is open. Now, some of you are chipping in with “but what about sensitivity?” Well, the sensitivity determines the brightness of the image from a given exposure. And since digital sensors enhance the brightness by adding gain to the photon count, usually in the analog domain but not always, then the apparent exposure (brightness of final image) includes the total amount of light with the gain applied.

So … who cares you ask? Well, as it turns out, I do. This weekend, a raucous debate on the Fuji forum centered around the concept of Equivalence. The idea of Equivalence being to achieve identical images from two entirely different cameras with differing sensors, lenses, pixel densities etc. The subject appears to involve more than a little mental masturbation, but there are many good points made in the text of the paper on which all the arguing was based. You are welcome to read all about it here.

Now … I was wondering how the light density would affect noise. The author of the paper swears that the only advantage of a larger sensor over a smaller sensor is the size of the aperture needed to get the same (equivalent) image. It goes kind of like this: you shoot the D700 at f/2.8 and the D300 at f/2.0 and you now have identical images. At least, that is how it should work since now the depth of field is the same and the diameter should be the same. I.e. we have matching densities of light. Actually, that’s not even correct … because if this is the same lens, then you need the aperture to be open to exactly the same size to get matching density … it is “total amount of light” that is equalized by changing apertures. See … it’s too complex to be of much use in practical photography. But it is a measurebator’s wet dream …

I want to explore some concepts from the paper … and I will over the next couple of articles. But let’s start by changing exactly one thing … the density of the light on the same sensor with the same apparent exposure. This has been a hugely controversial subject over the weekend.

So … leave ISO alone, since we want the same gain. Change the aperture by cutting it down dramatically, then increase shutter speed dramatically. Equivalent apparent exposures at different light densities (from the sensor’s perspective.) I set custom white balance as well, since auto white balance was changing at different exposures. That is a bit freaky by the way, as it indicates that the sensor is responding differently to halogen light based on shutter speeds.

Well, the F300 shocked me with how it responded. The longer exposure, despite being the same total amount of light, was extremely noisy in the blue channel. I t completely freaked.


Well, that’s just bloody awful. Clearly, with this camera (and I would be quite willing to bet that it generalizes fairly well with the tiny sensors), you want to avoid longer exposures if possible. I.e. you want to shoot wide open so that the light hitting the sensor is as dense as possible.

Update: It has been pointed out that 4s is in the territory of thermal noise. And of course this gets worse with the amplification of noise at higher ISOs. Still, the staggering size of the effect at a relatively short 4s exposure does not make a lot of sense to me as a purely thermal effect.

Update 2: I’ve been pounded on the Fuji forum by a group of individuals that I have dubbed the Gang of Four. There are actually two silent members and two obnoxious members who patronize the whole forum in their quest to dominate everyone in it. Interestingly, I’d come to think of them as a virus before finding out that they have done this in other forums. And like any virus, the nourishment they get allows them to keep going. I’ve stopped reading them at all. But meanwhile, I found this comment from Thom Hogan regarding long exposures, which I think puts in perspective my comments in this article. I had every expectation that 4s and 8s were not overly long exposures. Here’s Thom’s comment:

“Since the camera seems perfectly fine for long exposures measured into the low minutes, I'll take a longer exposure over higher ISO and keep the little bit extra in the shadows, just as I did with the D2x.”

Now … how did “low minutes” become “low seconds?” Ask Fuji I suppose … or maybe thermal noise is not the culprit the GOF would have us believe.

Now, 3200 ISO is a real stretch for small sensors, although you must admit that the left half is not terrible. In fact, let’s take a peek at a processed version.


Ok … it’s not great, but hey … you can’t expect much at 3200 ISO.

One more experiment for the evening. I lower the ISO to something a bit more reasonable and crank the lighting to allow me an 8 second exposure (max on the F300 remember) at minimum aperture of f/14. That gives me the required 3 stop range for the test.

And here is that test at 400 ISO … the difference is subtle enough in the noise area that I chose to present it as an animation instead of side by side etc …


Interesting how the smaller aperture picked up reflections. Unfortunate too … anyway, there is still an effect, and that’s good enough for me. Watch your light density … buy better lenses to keep exposures shorter at higher ISO. Especially as sensors get smaller … I’d be really careful of that with the 4/3 sensors, since there is definitely an option to shoot wider apertures.

Update: 8s is even further into thermal territory. Yet the differences here are much less obvious, no doubt because the amplification of the differences is lower. So thermal noise might be the culprit, but that does not obviate the conclusion that higher light density is good, which is all I chose to conclude from this experiment.

Next experiment will discuss the D300 and D700. How is the noise at the same light density? If I shoot the D700 in APS-C crop mode, then the only difference is the pixel density. 5mp versus 12mp in the same area on the D300. Sounds like fun … stay tuned …

No comments: