Noise reduction / sharpening / resize

Which order should this be done?

I assume sharpening should be done after resize. But what about noise reduction, before or after resize?

Generally you perform noise reduction as one of your first steps in post processing - before sharpening and before before image resizing.

Some noise reduction programs run quite slowly - and thus working with resized images may save time. However, the noise reduction algorithms built into programs like Neat Image and Noise Ninja need detail-free image regions in order to define a "noise profile." Working with the original image size makes more sense.

The basic point is to get the noise reduction done before you sharpen - unless you like sharpened noise/grain patterns (hey...they're no rules here).

I use PS only and I obtain great results with resized photos. I'm often preparing photos for offset printing at 20x28 inches/225dpi size from a 6MP file from my 300D. I use this sequence of actions:
1. Levels or curves correction
2. Despecle filter (obligatory)
3. If the photo is not very sharp I run Unsharp mask filter with radius 0.6-0.8 pixels, threshold 3 levels and amount 80-130
4. Resizing (interpolation) with the Bicubic smoother option selected.
5. Final unsharp masking with radius 1-2.5 pixels depending on the size of the final print. If the photo has some noticeable noise you may run a Median filter before the sharpening with radius 1-2 pixels and threshold 2-3 levels.
NOTE: This is not a painkiller receipe. You may have to do other things (actions) to make your photo as good as possible on the final print.

That depends on whether you are resizing up, or resizing down.

---Bob Gross---

I *always* run noise reduction first. My theory is that the "purer" the noise then the better the end result. Neat Image seems to be very well tuned to the cameras and their ways of producing noise, so I don't want to mess that up with any type of processing.

Again, this is just a theory I have. But I know that my edge-masked Neat Image NR process does a very good job on noise, so why mess it up?


Edit: Actually, I crop if necessary first BUT only if the crop won't remove a good sample area for Neat Image.

I've done Neat Image early, mid and late, and haven't noticed it makes any difference at all.

Maybe something is wrong. Either you shoot subjects much different from the rest of us, or you shoot with a low resolution, or something odd.

I found that (once I learned how to use Neat Image) I would get a serious improvement in 90% of the images. If the Neat Image-filtered version gets me no improvement, then I can delete that one, and the only thing that is lost is a little bit of time.

---Bob Gross---

No, the discussion was on the order of doing it. I use Neat quite often. The order is not critical, unless you have some scientific measurements I am not privy to. If so, please give us this info.

I agree. I can't see important differences between applying noise reduction early or late in the image editing process. I'm open to the possibility that there may be differences.

Neat Image (NI) and related programs reduce noise by optimizing the "wavelet" algorithms for:

-Different channels (typically, two channels for color noise and one channel for luminance noise)
-Different image spatial frequencies (amount of detail) – e.g. low, med, high, etc.
-Different image tonal values – e.g. shadows, midtones, highlights, etc.

If the noise characteristics are consistent across these variables in the image, then you should get good results. For example, red color noise in shadows with high frequency details should have the same characteristics wherever there are shadows for optimal results

Image editing changes image data for all of these variables. Thus a noise profile – if correctly set up – should differ at each stage of image editing – probably quite subtly in average cases.

I prefer to defer noise reduction until I have completed color and contrast adjustments - and after correcting tonality in the shadows. At that point, I can see the noise better (especially in the shadows) and evaluate what I'm doing in NI as I set up a noise profile. That's just my approach - probably no better or worse than others.

I can't see any, either.

I generally have to take the background sample early enough that I have not taken the final crop. Sometimes the final crop removes the piece of background that would have been a good sample for the filter.

---Bob Gross---