Noise Reduction in High ISO Nightscapes

[iceman (Mike)]

I use the Topaz Labs De-Noise Photoshop plugin for reducing noise in high ISO nightscape images.

This image shows the dramatic difference. Left side is straight out of the camera, untouched, from a ISO6400 nightscape image with the Canon 6D. The right side shows the same field after using the Topaz De-Noise plugin on the RAW Medium setting.

It's always the first step in my processing workflow.

Higher-res here: http://bit.ly/W48RSx
It's a 1:1 crop.

Also attached is the original image straight out of the camera, no processing except noise reduction. Shows how much light is captured at ISO6400!

Original image: ISO6400, f/2.8, 30s exposure

[gregbradley]

It certainly cleaned up the noise well. Its interesting most of the noise appears to be luminance noise rather than the usual chroma noise (smarties). That's half the battle already.

I find some noise reduction routines take out faint stars (Nikon View NX2 but their astro noise reduction isn't bad). Lightroom 4.3 is good for noise reduction as well, particularly chroma noise. I use Noise Ninja (but not so much on DSLR shots as it attacks the stars too much) but I think your Topaz is doing a better job there. Its a touchy area of processing as too much and it makes the image take on a plastic feel as it is basically blurring. Too little and it looks grainy. Selective noise reduction is the usual algorithim that works best where it is applied mostly to dim areas and not bright areas.

By the way have you experimented with what point you do noise reduction? With DSLRs I do it with RAW conversion and perhaps late in the processing chain if it needs more and with CCD it is done early (deconvolution usually on the combined masters of each LRGB - usually just on luminance but sometimes on one or all of the RGB to get star sizes to match).

I'll download a trial of it and check it out more.

Greg.

[colinmlegg (Colin)]

You may want to try a temporal NR method and compare results.

I use Neat Video interframe (temporal) and it tends to work better than the intraframe option ... less smoothing. Interframe requires 2 or more similar images taken a short time apart (usually from a timelapse).

[DavidTrap (David)]

IMHO, you've gone too far. Plastic and blurry are the right words.

I've had best success with Lightroom's noise reduction, although I've never tried the third party stuff.

DT

[LewisM]

I tend to agree with David - it's lost a lot of impact now, and does look artificial. Worse is Nik's DNoise, but I find if I use Focal Pointe Observatory's "Select Stars" action first to isolate the stars, and then DFine, with 55% or LESS fill, it does a MUCH better job.

[iceman (Mike)]

That's the good thing about Topaz - it has a range of different presets, as well as the ability to customise your own settings. So you can choose how much or how little you want to apply.

[gregbradley]

Mike have you considered shooting in jpeg after experimenting with the ideal white balance temperature?

One of the 6D's strength is superior noise reduction in camera in jpeg.

I know its sacrilege but if you know the ideal white balance then shooting in RAW does not add that many advantages over high quality jpeg. I've already sussed out the best settings for my D800E so I don't need to fiddle too much in post processing RAWs.

6D has in camera RAW conversion (so does Fuji XE1). So perhaps you can access the jpeg engine that way and still shoot in RAW.

Greg.

[iceman (Mike)]

It's not just white balance though Greg.

Jpeg is 8 bit. Raw is 14 bit, so there's a lot more control to bring back the shadows and highlights than if you shot in JPEG.

Also, with all the saving and re-saving of images I do during processing, the first thing I'd have to do is convert the JPEG to TIF anyway, so there's not much point in capturing in JPEG.

[Phil Hart]

I've been having another play with Colin's temporal noise reduction process and have finally 'seen the light' after some not very convincing attempts previously.

In Adobe After Effects, I have the Neat Video plug-in. You have to Profile the image as per normal first but then turn the in-frame noise reduction to zero. Then set the temporal noise reduction 'radius' to 1 or 2 (means it is using one or two frames on either side in the sequence for comparison). Even with a setting of 1, it is effectively averaging three frames together and makes a huge difference. Increasing it to 2 doesn't yield a huge amount more, and above that the processing time starts to increase a lot for little gain (might be more relevant in a real 'video' file in other situations) and you need a longer sequence of images as well.

The example attached is with a temporal radius of 2 and a threshold of 250% which I found better than lower threshold settings . [Original is on the right]. This does a great job of reducing noise while maintaining detail - note this is effectively like image stacking so there is a real gain - not just noise smoothing. (Some colour noise reduction was also applied in Lightroom first but no luminance adjustment).

Phil

[colinmlegg (Colin)]

Yep, that's pretty much what I do Phil. More than 1 either side tends to smooth more due to movement in the image. I've tried a couple of other temporal plugins and neither was as good as Neat Video. I've even heard some professional editors saying it's the best out there.

So the little secret is officially out

[gregbradley]

What a great thread.

So to use the Neat Video plug in you need at least 5 images of the same scene? For a time lapse I can see you would have lots of images to apply but for a single nightscape you'd have to take 5?

If so would you get the same result by simply stacking those 5 images?

Its like a dithering approach.

Greg.

[colinmlegg (Colin)]

Originally Posted by gregbradley

So to use the Neat Video plug in you need at least 5 images of the same scene? For a time lapse I can see you would have lots of images to apply but for a single nightscape you'd have to take 5?

If so would you get the same result by simply stacking those 5 images?

Its like a dithering approach.

Greg.

3 (total) is good enough Greg, even 2 has worked for me in the past. The plugin does more than just stack. It has motion detection software, so can work out what has moved in an image (to some degree) and apply the noise reduction in an intelligent way.

With normal stacking (DSS) you have the issue of moving stars vs static foreground (or vice versa). Either you live with some blurring in one or the other. Or stack twice and blend the 2 results.

[gregbradley]

Thanks Colin. Very helpful. I'll have to try it out.

Greg.