More details are on my flickr stream starting with this image.
Image with no noise reduction at all. This is extremely noisy because I pushed the shadows quite a bit to counteract the backlighting in the image. I did not use flash in this image, as I think it's kind of mean to the animals. Other people may not have the same moral objection to using flash, and may use flash freely. If it's one more bright flash the animal doesn't have to endure, then all the better.
Using the built-in noise reduction in Adobe Camera Raw. This is the most you could ever hope for in ACR. Better, not not much better:
Noise Reduction using NeatImage in an automated workflow. The chromiance noise has been eliminated, but the luminance noise is quite strong. This is mostly due to NeatImage's poor choice for a profiling zone. It chose the white area in the background of the image, which does not exhibit much noise at all:
NeatImage in a manual workflow. Essentially, the PS Script stops at the noise reduction step allowing the user to intervene, allowing the noise reduction algorithm to get optimal results. By choosing a more representative area of noise (in this case, the bird's chest does not contain much detail, so is a good candidate) Simply by choosing a better area to profile from improved the image greatly, but left some speckling, so tweaking the high-frequency and luminance noise levels, I was able to virtually eliminate all speckles left over from the noise reduction process, giving a much cleaner final image:
All images were processed using a batch script, only changing how noise reduction was applied. As part of this processing, sharpening using the Smart Sharpen filter in Photoshop is applied after downsizing to help counteract some of the softening that normally occurs when the image is resized.
The drawback, of course, is any noise reduction algorithm will attack details, so a balance must be struck between acceptable noise levels and noise reduction artifacts (e.g. Plastic appearance, softening, and loss of color saturation)
There are other choices for adaptive noise reduction available. Any adaptive noise reduction package will work. I have used NeatImage for a few years, and prefer that package over others. The key is to choose a package that uses an adaptive algorithm (such as Neat Image) these algorithms will require a profiling step using a sample from the image (or another image) to calculate the exact nature of the noise produced by the camera. It should allow the flexibility to adjust the reduction and estimation levels in separate channels, and separate frequencies to optimize the amount of nose that will be removed.
There are trade-offs of course. You will need to balance the amount of detail that will be retained to the amount of nose that will remain in the image. Even the best algorithms will have this trade-off.
I hope this post gives you some insight into reducing noise in your images. Comments are appreciated, and feel free to post questions as well.