Noise reduction with CS2
rutt
Registered Users Posts: 6,511 Major grins
I'm jumping the gun on Margulis Ch 5 (sorry XO), but I've been testing out Dan's ideas from that chapter. In short, I found I couldn't just apply Dan's recipe for noise reduction, but I have had some good luck with an alternative. I'm interested in getting some feedback.
I'll be working with this picture which I shot yesterday with a Canon 5D at ISO 800 at Fenway:
Original: here
Davev suggested a great crop which really helps the picture tell its story. This is my version of his crop:
Original: here
But now the noise is very noticeable. Davev used NoiseNinja on this image when he suggested his crop (see below). I've been reading Dan Margulis' Photoshop LAB Color, Chapter 5, so I thought it would be a good opportunity to try out a technique presented there.
Dan opens the chapter with an image with a lot of color noise. Converted to LAB, his image has little noise in the L channel and quite a bit in the A and B channels. This up and right down the middle for Dan, who shows how to blur the A and B channels, basically losing nothing but the noise.
When I tried this, I found the A and B channels pretty clean and the L channel had all the noise. Bummer. Back to the drawing board. (Actually, I've encounterd this same problem before, since Dan has been advocating blurring the A and B channels for years. And for years, Canon cameras + ACR have made images with L channel noise and clean A+B channels. I think this may be evidence that Dan still has a foot in the scanner world. I've tried to capture his attention with this issue before and now I'm trying again. He's likely to care more now that he is working a next version of Professional Photoshop with an emphasis on digital photographers.)
One thing I did learn from reading Ch 5 is that PS/CS2 has this nifty new blur tool, called "Surface blur". It works a little like unsharp mask. It can detect edge transitions and only blur where there are none (hitting them where they ain't.) Unlike USM, increasing the value of Threshold in this filter increases the amount of blurring it does. The threshold value tells it how big an edge transition to ignore. So setting the value very low means there will be no blurring where there are even subtle edge transitions. 2 is the lowest possible value.
I blurred my image's L channel with Threshold=3 and Radius=5. A Gaussian blur with a radius of 5 would destroy the fine details in this image, for example the hairs and skin texture. But surface blur did a very creditable job here:
Original: here.
I followed with the usual LAB curves and separate lighten and darken USM and ended up with this:
Original: here.
I think this is at least competitive with what Davev got with NoiseNinja:
and has the advantage that I actually understand what it's doing.
I'll be working with this picture which I shot yesterday with a Canon 5D at ISO 800 at Fenway:
Original: here
Davev suggested a great crop which really helps the picture tell its story. This is my version of his crop:
Original: here
But now the noise is very noticeable. Davev used NoiseNinja on this image when he suggested his crop (see below). I've been reading Dan Margulis' Photoshop LAB Color, Chapter 5, so I thought it would be a good opportunity to try out a technique presented there.
Dan opens the chapter with an image with a lot of color noise. Converted to LAB, his image has little noise in the L channel and quite a bit in the A and B channels. This up and right down the middle for Dan, who shows how to blur the A and B channels, basically losing nothing but the noise.
When I tried this, I found the A and B channels pretty clean and the L channel had all the noise. Bummer. Back to the drawing board. (Actually, I've encounterd this same problem before, since Dan has been advocating blurring the A and B channels for years. And for years, Canon cameras + ACR have made images with L channel noise and clean A+B channels. I think this may be evidence that Dan still has a foot in the scanner world. I've tried to capture his attention with this issue before and now I'm trying again. He's likely to care more now that he is working a next version of Professional Photoshop with an emphasis on digital photographers.)
One thing I did learn from reading Ch 5 is that PS/CS2 has this nifty new blur tool, called "Surface blur". It works a little like unsharp mask. It can detect edge transitions and only blur where there are none (hitting them where they ain't.) Unlike USM, increasing the value of Threshold in this filter increases the amount of blurring it does. The threshold value tells it how big an edge transition to ignore. So setting the value very low means there will be no blurring where there are even subtle edge transitions. 2 is the lowest possible value.
I blurred my image's L channel with Threshold=3 and Radius=5. A Gaussian blur with a radius of 5 would destroy the fine details in this image, for example the hairs and skin texture. But surface blur did a very creditable job here:
Original: here.
I followed with the usual LAB curves and separate lighten and darken USM and ended up with this:
Original: here.
I think this is at least competitive with what Davev got with NoiseNinja:
and has the advantage that I actually understand what it's doing.
If not now, when?
0
Comments
Just in the nick of time!
I was JUST about to start PS-ing some heavily noised ISO1600 pikchas from last nite. Your post would help a lot!
I have already discovered the hidden power of the surface blur for meself.
It's a great tool, but thus far I didn't used it in LAB, silly me..
I will try to follow your lead and post the results as soon as I have them:-)
Cheers!
Make sure to check the A and B channels first. If the noise is there, you're into Dan Margulis Ch 5 land and you won't have to be nearly as careful to get better results.
I wish I could try that lab stuff, but PSE 2 doesn't have lab in it. Or at least I
haven't seen it.
The one step that I did that you didn't was to unsharpen mask the phone screen
after I used noise ninja. With what you did, and making the screen just a hint
sharper, you've got a great photo.
P.S. you gave me way to much credit for your story telling picture.
Thanks.
Basking in the shadows of yesterday's triumphs'.
Portfolio • Workshops • Facebook • Twitter
Andy, that's a darn good point. I was just reading the chapter this and then DaveV ran this image of mine through NN so I thought I'd play. When you have a hammer, everything looks like a nail.
Still, I'd like to see what people get with images with really bad noise. I guess it can partly wait for XO's post on Ch 5, next weekend he says.
Does anyone know where the best supplier would be to get it for a Canadian Photog?
http://framebyframe.ca
[Bodies] Canon EOS 20D - Canon EOS 500
[Lenses] Sigma APO 70-200 f/2.8 - Canon EF 85mm f/1.8 - Canon EF 50mm f/1.8 - Tamron XR Di 28-75mm f/2.8 - Canon EF-S 18-55mm f/3.5-5.6
[Flash] Sigma EF500 Super DG Flash
[Tripod] Manfrotto 055 Pro Black
[Head] 484RC2, 200RC2
Got mine at Amazon. Barnes & Noble has been known to stock in local stores. The book is a best seller, so it's not hard to find.
And once you do get it, how about writing a chapter summary?
http://www.moonriverphotography.com/gallery/855603/1/38499428
Here are the L, A, and B channels of the ISO 3200 crop:
I used surface blur on the L channel of this (where the noise is), tuning the parameters for the 5D (noiser 1Ds Mark II will require both higher threshold and radius.)
Here is Andy's original (left) and my surface blurred version (right):
OK, let me sum up the technical part of this discussion to date.
Noncontroversial: Modern Canon cameras make much more lumonisity nose than color nose at high ISO settings.
Experimental: Surface blur of the L channel early in the work flow is a good solution for this problem.
Open: A much better solution. Noise Ninja, Neat Image, or the like are nice, but I would prefer to have a lower level more transparent solution which would actually contribute to my understanding.
(Weird statement).
I like the lack of noise on the 5D.
(That's better)
Dgrin FAQ | Me | Workshops