Eliminating Color Artifacts

[DBrigleb]

If this is common knowledge in the Photoshop world, I apologize. Although I've been in and around photography most of my life, I'm new to digital and PS.

To get rid of the color noise that appears in underexposed digital images I do the following:
1) Make a duplicate layer of the image.
2) Do a Gaussian Blur of the dupe. Use enough blur to eliminate the noise spots.
3) Set the dupe layer's blend mode to 'Color'.
That's it.

I've attached a small PS file to demo the procedure. You'll probably have to zoom in to see the effect.

[DJ Dubovsky]

Hi Dick,
Thanks for the cool tip. I have to try that one out. Oh, for some reason the file didn't follow your post. You might want to check the size and resubmit it.

I was taught another way to deal with color noise especially in digital photos by changing it to LAB mode and adding a soft blur to the A and B channels. I will have to try yours and see if it may be a better way. Thanks again for the great tip.
DJ

[Stephen M]

Hi Dick, thanks for sharing the tip (what is common knowledge for some may not be for all - so it is good to spread the good stuff around. <g>)

Similar to this previous thread:

http://www.retouchpro.com/forums/showthread.php?threadid=4113

P.S. Attachment paths seem to drop off the interface when you preview your post...if you preview they must be reentered again? At least this seems so on Win98/IE5...should this be reported as a bug to Doug?

DJ - Yes, LAB is great in that it separates the more sensitive data (A red/green) from the less sensitive colour components (B blue/yellow)...so you can often afford to hit the B a little more than the A channel...but -

The mode change to LAB will result in the loss of some minor data - even if the eyes don't see it...it still happens.

So then there are three choices:

i) The benefits outweigh any disadvantages, so you go to LAB anyway

ii) You go to LAB but do it in a dupe, then reblend this dupe into the original file in color blend mode, so that the original images luminosity component is not affected by the LAB move (we are only using this technique for colour and not tone, so the whole image does not need to go to LAB...)

iii) Or you decide that it may not really matter that you can have more control by having split red/green or blue/yellow filtering control - and you just dupe a layer and set it to color blend mode while in RGB or CMYK modes instead of using LAB.

All of the above is in a small article of mine, as well as the introduction of edgemasking into the hue/chroma artifact and noise reduction...this way high contrast edges are protected from any noise filtering.

http://members.ozemail.com.au/~binaryfx/IT_artifactremoval.html

There are also related links to more info at the end of the article.

Another method is to use the history brush set to color mode to paint back in the uncorrected hue/saturation data to a color filtered file. This is done to fine detail and other areas that may have become desaturated or have hue shifts due to the colour component filtering.

I like despeckle, median, g/blur and smart blur as possible noise reduction filters...run in smaller amounts and combined rather than just using one bigger blur, but it all depends.

Regards,

Stephen Marsh.

[DJ Dubovsky]

Thanks Stephen. That's a good tip and an excellent alternative to losing data switching modes. I may take that example to other situations where I would be tempted to switch modes for certain procedures.
DJ

[Stephen M]

Don't let it worry you too much DJ - as I said it can be rare that it makes an actual difference to visual or editing results...but if it does you do have an alternative.

More on the topic can be found here:

http://www.ledet.com/margulis/ACT_postings/ACT-LABLoss.html

So thus I use luminosity or color layer blends/fades etc to try to simulate what LAB does without leaving RGB or CMYK where possible, but there are times where LAB is better for certain tasks and the mode change matters not, as the benefits gained from the trip to LAB are worth it.

Stephen Marsh.

[Guest]

DBrigleb, this is a very good tip. I first learned this from an online tutorial somewhere (?) and have found it very useful. That version was a slight bit different in that instead of gaussian blur it called for noise-->median. I suppose anything that blurs the colors together would work however.

Phyllis

[DBrigleb]

PStewart,
I've used median as well. It gives better definition between color area.
Also, I the subject of LAB, I thought I read somewhere that when you convert from one colorspace to another Photoshop goes through LAB for the conversion.

[Stephen M]

There are two different forms of LAB, when it comes to Photoshop.

The 'mode' that you can edit in is 'ICC LAB' and it is not as accurate as a true 'CIE LAB' space...but may be hard to visually see the difference anyway, depending on the original and the transform.

When you use a convert to profile command - Photoshop uses source and destination profiles...which go through a 'Profile Connection Space' or PCS, which is either CIE LAB or CIE XYZ. This is not the same as editing in Photoshop's LAB mode (the mode transform also uses 16 or 20 bpc math behind the scenes). The standard mode change to ICC LAB mode is usually done on 8 bpc data.

Back in Photoshop 4 (not sure on 5.x), then yes - LAB was the behind the scenes space that served the same purpose as a PCS...but I would hope that the internal LAB used between mode conversions was CIE and not ICC.

P.S. Some older Linohell scanner software uses a form of LAB which has a different whitepoint than the Photoshop LAB. But for any version of Photoshop on any platform, the Photoshop LAB mode is the same between all versions.

Regards,

Stephen Marsh.