As great as digital camera sensors have become at combating noise, this film-grain-reminiscent interference can still rear its ugly head at high ISOs and in long exposures. So, removing excessive noise—literally improving the signal-to-noise ratio—is as important as ever. There are plenty of traditional approaches to eliminating noise—from using Photoshop's Reduce Noise filter to standalone programs like Noise Ninja and Neat Image, as well as RAW converters and their noise reduction features—but all of them risk compromising the image quality by removing actual image-forming detail. That's the real challenge for any kind of noise reduction—separating the signal (the image-forming detail) from the noise.
Since noise is random and changes with every shot, it makes sense that if you made multiple exposures of a scene and all of the actual image-forming detail remained the same from frame to frame and the noisy pixels would be different in every frame, you could analyze the exposures as a group to better determine which pixels comprise image-forming detail and which ones are just noise. That's the idea behind averaging multiple exposures to first identify and then eliminate noise.
To start, shoot multiple RAW exposures of an unchanging scene that would benefit from significant noise reduction. (RAW is useful because of the additional random pixel changes that occur with JPEG compression.) Then, open those exposures—let's say there are five of them—into separate layers of a single Photoshop file. Then you simply adjust their opacities so that they are all equally weighted in the final image. Think about it: if you set one layer's opacity to 50 percent, you would effectively remove 50 percent of the noise from that layer, and where the image-forming pixels match the image-forming pixels of the layer below, they would stay the same. Better still, where the image-forming pixels of that layer overlay noisy pixels from the layer below, you'll reduce the lower layer's noisy pixels by 50 percent too. Neat, right?
What you can't do, though, is simply pile layer upon layer and set each one to 50 percent opacity. You'd waste their noise-reducing power that way. But, you also can't just determine a layer's opacity by dividing 100 percent by the number of layers—in this case that would mean each of the five layers would be set to 20 percent. In actuality, you need to divide a layer's opacity by the number of layers below it plus one. So, first set the base layer, the image background, to 100 percent opacity. The second layer would be 50 percent opacity (100 percent divided by 2). The third layer would be set to 33 percent opacity, the fourth 25 percent, the fifth 20 percent, and so on. Zoom in to 100 percent detail and compare this composited image with a single noisy exposure, and you'll not only see less noise, but the overall appearance will be noticeably crisper and clearer. You'll have reduced noise without compromising any of the image-forming detail. A major feat.
The real caveat, of course, is that all of the image-forming pixels—the detail you want to keep—must remain the same from shot to shot. This means you're limited in the type of photographs this technique will work for: still life, architecture, very still landscapes—anything that doesn't move at all. But, when you've got a shot that qualifies, why not make multiple exposures and layer them together in the computer to help reduce noise without compromising detail.