Our friend Eolake Stobblehouse wrote to Ctein and me this morning with a couple of links to Imaging-Resource and the following observation:
What made me try Noise Ninja on the uncorrected sample was not so much the noise itself as the hash the camera-correction had made of correcting the picture: it was 'mushy.' Look at the olive oil bottle, both the outline and the rendering of the oil inside. It's not attractive.
I was shocked to discover how poor a job the camera does with noise reduction compared to the computer. Maybe this is well known to some, but I am always reading about cameras and noise and so on and I had no idea, so it seems to me that this aspect is very under-reported.
Well, I can only bleat, plaintively, that I try to report it. (What I mean is that I bitch about it from time to time, and am roundly ignored.) First of all, forum denizens who carp about noise and don't have Noise Ninja or Neat Image or FixerLabs NoiseFixer (Photoshop does a pretty good job itself now) or one of the other good noise-reduction apps or plug-ins are being foolish (note, that's not ad hominem: I didn't say they were fools, I said they are being foolish). What you do is, you run a few tests with your DSLR at various ISOs, run your anti-noise program on them, decide at which setting either the noise or the noise-reduction artifacts start to seem a bit obtrusive to you, and then stick to one setting lower than that unless faced with extraordinary circumstances. If that "last good setting" is too low, buy yourself a better camera that does okay at a higher setting.
I have never understood the obsessive fascination with noise. It used to be a problem, granted. It's still a problem with some small-sensor cameras—okay. But digital SLRs are very good now. I suspect most other film-era holdovers don't get it either—we had to get used to grain, and grain was worse. (I do remember being allergic to grain, in the beginning. I learned to live with it. To love it, even.) The fact is, I don't mind noise. It just doesn't get in the way of seeing the picture, at least with the kind of pictures I take. Sometimes it's even pretty. I suspect that some "newbies" are hyper-sensitive to it because it's not a feature of the world in front of their cameras and they're affronted by traces of it in their pictures. Okay, but that's only partly an actual problem—the other part is in their heads: it's that they need to get over it.
In spite of all that, noise reduction software is really, really good. Really. So my advice is to buy one and learn how to use it, like Eolake has. And you'll be just fine. Noise really is one of the biggest non-issues in all of digital photography, IMNSHO.
Ctein may wish to chime in here, and if he does, I'll publish his thoughts just below.
Mike (Thanks to Eolake )
Ctein adds: Dear Eolake, Nothing surprising, nor (I don't think) any kind of secret. The camera's trying to do, in a small fraction of a second, with less than a watt of power, what your computer is taking many seconds or minutes to do, consuming tens of watts of power. Even allowing for dedicated signal processor chips in the camera, there's just no way the camera can throw anywhere as much processing power at the problem.
The amount of internal signal processing and computing in the camera makes a huge difference in image quality. Look at my latest column reviewing the Fuji camera, and look at how much difference there is in the JPEGs between the S6000 and the S100, even though the S6000 has an inherent advantage in raw sensor noise; entirely due to computing power.
Featured Comment by Thom Hogan: "Ctein, I think you're a bit off the mark here. Dedicated imaging ASICs are every bit as powerful as a computer-based approach. Indeed, DIGIC and EXPEED both use the same underlying code and algorithms as their desktop equivalents, and are faster. That's why you create ASICs. The difference is in flexibility. If you don't like the regular profile you can create your own with Neat Image, et.al. Indeed, if you shoot slightly off in exposure, shoot in slightly different light balances (mixed light, especially), and a host of other small differences, then you may be better off using raw and a third party noise reducer. But for the situations they were designed for, a good in-camera ASIC is going to be as good as using that same code externally in a computer, and faster to boot.
"One issue that comes up is when the ASIC was designed and how often it is iterated. Companies like Canon and Nikon get advantages in terms of scale. When you're shipping fewer units you have a tough choice: eat the development cost over fewer units (e.g. higher product cost), or iterate less often."
Featured Comment by Adam Isler: "Just yesterday I was processing some high ISO photos taken with my Nikon D300 in the New York subway. I had used the adjustment brush in LR to boost the exposure on some faces and was erasing some of my clumsy initial strokes so had zoomed way in. And I was thinking how lovely and film-grain-like the noise was.
"And then I stopped and had to wonder how and why I thought like that. I mean, it's not like film grain is 'natural.' But that's how we act. Like somehow film grain is an organic, all-natural product but digital noise is some nasty, artificial pollution. I think there's a nostalgic circuit in our brains that operates below the level of consciousness.
"Last summer I tried shooting with my Minolta 7D, boosted to its highest ISO setting, trying to get a pointillist effect, to no avail—just couldn't get enough noise—but I didn't want to just use Photoshop to create the noise—that felt like it would be cheating. Another example of some bizarre cognitive activity imposing utterly meaningless constraints on picture taking and picture making...."