Noise removal...Before, or after ?

[cmM]

Do you do all the post processing first, or the noise removal first ?

I've never dealt with any noise removal software, but last night I was forces to use ISO-800, and I think I'm going to need it.

[maderito]

I don't think there is any *right* answer to your question. What follows is my approach.

I use noise reduction programs (Neat Image, Noise Ninja) which rely on algorithms that first characterize the noise before reducing it.

I do global edits first - color caste removal, setting white/black point, adjusting contrast/brightness, and fine tuning shadows and highlights.

At this point, noise will have approximately the same visual appearance and the approximate statistical features that will be present in your final print. So, I next run the noise reduction program, perform final editing tweaks, and then sharpen for printing or web display.

Depending on my mood, sometimes I remove the noise earlier during editing - especially when there's a lot (e.g. ISO 1600 or more). When there is little noise (e..g. ISO < 800), I sometimes remove noise as one of the final editng steps on a duplicated layer and then merge, adjusting opacity to taste. This works well when you're dealing with textures that need a touch of softening - like baby's skin - and when the noise varies in the mage, something the noise reduction programs handle better than a global gaussian blur.

Mostly, I don't worry about noise. 8)

[Scottes]

Noise Removal is the very first thing that I do. My belief is that any type of manipulation or processing will change the noise - possibly pushing some of it to a point where it isn't recognized as noise.

However, I have no basis for this, just a belief.

So, as I'm wont to do lately, I ran a test. I clipped a section of an ISO 800 image containing detail and an area with a lot of noise. I performed a number of steps on each, identically. Except that on one I did NR first, the other got NR last. (Using Neat Image.)

I couldn't tell the difference zoomed at 300%.

However, I would strongly suggest runnning NR before sharpening or resizing at the very least. Either of those two would definitely screw up noise reduction attempts.


As a note, there are time when running NR before cropping is very beneficial. I use Neat Image which requires an area of the image to be analyzed. I've found that I get much better NR after analyzing an area containing both dark and light sections. And I've often found that I crop out areas with a bright reflection that doesn't belong, or an out-of-focus branch that doesn't belong, or some other such disturbance. That bright spot on a dark background is a great area for analyzation, so I may use that for the NR then crop it out after NR is done.

[JZaun]

Ok this seems to be the place to bring this up... I hear all the talk about noise reduction programs., I dwnloaded Neat Image and tried it. Worked well.. I thought.. :? today I took a image I had noise reduced with neat image and used PSE2 /filter/noise/despeckle.. It cleared up the neat image job.. If PSE2 is so good why should I use neat image...My results with the noise filter in PSE2 hgas been very good. I can apply it to the whole image or block out a part of the image to use it. I never hear of anyone using the PS filter. Am I missing something ?? I know I got lots to learn and asking is HOW :D

JZ

[robertwgross]

As a note, there are time when running NR before cropping is very beneficial. I use Neat Image which requires an area of the image to be analyzed. I've found that I get much better NR after analyzing an area containing both dark and light sections. And I've often found that I crop out areas with a bright reflection that doesn't belong, or an out-of-focus branch that doesn't belong, or some other such disturbance. That bright spot on a dark background is a great area for analyzation, so I may use that for the NR then crop it out after NR is done.

I could not agree more. I also use Neat Image, and I also tend to crop out parts of the image first. I often scan around the image looking for "signs of color noise" and then use that sample, if possible.

I prefer analysis to analyzation.

---Bob Gross---

[cmM]

well in this case if the results are close enough, I'll just use photoshop's filter. I'll see how it comes out.
I'll post pictures later.

Thank for the info :wink:

[who10]

I use Neat Image which requires an area of the image to be analyzed. I've found that I get much better NR after analyzing an area containing both dark and light sections.

I have changed my workflow to incorporate the NeatImage test target. At first it seemed like a lot of extra work, but the results are exceptional and I no longer need to "find" featureless areas to sample in my pics. I first sort NR candidates by ISO/aperature (using Extensis Portfolio), create NR profiles for each pair as needed and then load the corresponding profile to batch process all the images in each group at once (for example all ISO800 at f4). It takes less time to create a new profile than it does to run one analysis on a single image - you simply bring up the test target on your monitor, focus at infinity filling the frame with the target, set camera ISO/aperature - take a picture of the screen and then run NR saving the resulting NeatImage noise profile for later use. This method virtually eliminated instances of NR confusing image detail with noise.

And I run NR immediately after RAW conversion, prior to Photoshop processing of any kind (using C1 as my raw processing tool).

David

[cmM]

David,
thank for all the great info.

In C1, do you sharpen at all of just leave everything to 0 and do it after (in PS) you remove the noise ?

[who10]

David,
thank for all the great info.

In C1, do you sharpen at all of just leave everything to 0 and do it after (in PS) you remove the noise ?


cmM,

I process the "target" with the same C1 sharpening settings I use for my images (generally default sharpening with zero C1 noise reduction). In the couple of tests I've run this gives me the best base image to begin Photoshop work. An interesting experiment would be to apply all of the C1 preprocessing (sharpening/C1 NR/levels/curves/whitebalance,etc) used on a particular image to the test target then creating a profile from that base. I wouldn't do that as "standard" workflow, but it would be interesting to see the affect...

As a side, I sharpen at least twice no more than three times (initial capture sharpening to restore sharpness muted by the 10D always in C1, sometimes some creative sharpening in Photoshop to obtain a specific affect, and final output sharpening - also in Photoshop - tuned to Web or printer requirments).

David