Noise Reduction Plugins?

[ktuli]

I tried searching for this, but was not very successful, so please forgive me if this is a duplicate thread.

I recently took a good number of photos at high ISO (both in the Monterey Bay Aquarium and at night in Yosemite) and I really need to do some significant noise reduction on them before they would be presentable.

I know someone mentioned using the Neat plugin, but the free version only works on images up to 1024 on the longest edge.

What noise reduction plugins or methods do folks use on their high ISO images? I really want to share these images, but right now, all I see is the noise!

Thanks!

- Bill

[Polar01]

Have you tried using noise reduction in the raw converter, it would help to know what you use to do you post production in.

Cheers:

Allan

[tclune]

I used to use Neat Image for noise reduction, and always rather liked it. A few months ago, I became interested in standardizing on one company for my plugins as I decided to switch to 16-bit image processing (my Neat Image was the cheaper license that was only 8-bit). So I switched to Topaz DeNoise 5 and I've been very favorably impressed by it. In addition, I use and like Topaz Detail 2. However, I was going to also use Topaz InFocus, but never really got the hang of it. It just doesn't work in my hands to anywhere near the degree of my old stand-by FocusMagic, which is still 8-bit only. So my migration to one family of plugins has foundered for now, but I can readily recommend Topaz DeNoise 5. FWIW

[ktuli]

Allan,

Sorry, I should have included such useful information, huh? - I am using Photoshop CS5 with RAW files, so I do run everything through ACR first.

I'm not very knowledgeable at doing PP work, so if you have recommended methods that work just in ACR, that would be appreciated too!

- Bill

[DanK]

I don't know if the NR controls in LR are available in ACR, but they are quite good.

I have used Noise Ninja for several years when LR is insufficient (which is not all that often anymore). I have been pleased with it. I picked it because a comparison of some of the leading software several years ago showed them reasonably similar in NR, but Ninja did less damage to color balance and saturation. However, that was some years ago.

[Carl in Louisiana]

I use LR4.1 and Nik's Define 2.0 and I am very satisfied with it, it also works with ACR and there is a version that will work with both ACR and CS5 if I am not mistaken, but don't hold me to that, check their web site to be sure. You can download a trial version for free to test it out. Okay I went and picked up the link.

http://www.niksoftware.com/dfine/usa/entry.php

[Poaceae]

Like Carl, I use LR4 and Nik Define. I find the noise reduction in LR4 to be very good - clearly better than that in LR3.

[CBImages]

Another noise reduction program you should consider is from a company called Imagenomic and is called Noiseware, it is excellent!
I have tried many of the other NR programmes and they all do a good job but for my money Noiseware has the edge.

As has been suggested before, use the NR tools in your RAW editor first on a fairly low setting before importing into Photoshop and then running whatever NR software you choose to go with. Always run the NR software on a duplicate layer - as you can then blend the effect using masks or the opacity slider to retain detail in selected areas.

[kdoc856]

Chris,

Just curious. Why take the extra step of running a low level NR in ACR if one is going to utilize the plug-in? Are their algorithms sufficiently different to a render a qualitative rather than just a a redundant effect?. I also use a NR plug-in, but have just gone straight to it. Am I missing a good thing here? Thanks

Kevin

[tclune]

Why take the extra step of running a low level NR in ACR if one is going to utilize the plug-in? Are their algorithms sufficiently different to a render a qualitative rather than just a a redundant effect?

Let me take a run at replying to this. There are basically two mathematically different approaches to noise removal -- convolution kernels and selection kernels. These classes of algorithms will largely not interfere with each other. What a selection kernel does is define a region that is used to evaluate the pixels non-linearly. For example, a median filter identifies a group of pixels that should be organized from smallest to largest, and the routine then chooses the one in the middle of the line-up as the "result" of the median filter. If you use a 3x3 square median filter, the routine looks at all neighbors of a pixel -- left, right, up, down, and sideways as well as the pixel itself. It then replaces the pixel with whichever of those is the middle value. What this routine is good for is getting rid of salt-and-pepper noise.

The other main class of filters, convolution kernels, are all linear filters. The filter coefficients define a frequency rolloff to apply to the desired two-dimensional region. Commonly, convolution kernels are larger than selection kernels. For example, if you are suppressing noise, you would use a low-pass filter of some sort. This will look at a region around the given pixel and smooth out high frequency data (fast changes) while accentuating low-frequency data (slow changes). The net effect of these kinds of filter is to make the image look a bit out of focus, but with less noise. When you apply a low pass filter and then a sharpening filter, you are essentially working at cross-purposes with yourself. First, you're throwing away high frequency data to suppress noise, and then you're high-passing the data to try to recover at least the appearance of sharp focus. One way that many programs avoid losing sharpness is to allow you to low-pass just the chrominance data, for example, and let the luminance data hold onto your sharp edges for you, so you have only blurred part of the image data. The particular noise in a given image will determine whether this approach is appropriate or not.

You would often be better off to apply a median filter to suppress noise and then high-pass to bring out sharpness in the image. Your two filters are then not fighting each other. But a median filter may or may not allow you to get rid of the particular noise you see in your image, so the actual choice of filters to use will depend on their effectiveness for your purpose -- a median filter that suppressed the "pilling" that you often see in high-ISO image noise would have to be huge and would result in posterizing your data.

Equally important to the underlying math in a filter is the user interface. Lots of noise reduction programs do essentially the same thing, but the ease of use will vary wildly between programs. My choice of tools usually ends up revolving around how easy it is for me to figure out how to apply the kind of process that I think an image needs more than that the given tool supports the kind of math I think is called for.

All of this is a long way to go to say that the plugins may well be the same thing that you could do with the native program itself, but the plugins may be easier to control. Under almost no circumstances would it make sense to apply both to your image. FWIW

[hoffstriker]

Great break-down Tom!