Sharpening in LAB Mode

[Fabs Forns]

One way to avoid halos and artifacts is sharpening your files in the Luminosity channel in LAB color.
Quite easy to do.

Go to Image>Mode>Lab color
Go to the Channels in the Layers palette
Select the Luminosity Channel. Your image will look black and white
Use Smart Sharpen at 100%, remembering, the bigger the file, the more sharpening you'll need.
Apply the sharpening.
Got back to Image>Mode>RGB color

[Robert Amoruso]

Nice work. I started doing this a while back for the reasons you discussed. Thanks for posting this tutorial.

[Joerg Rockenberger]

Does this method still require selective sharpening of the primary object only?

Thanks, JR

[Fabs Forns]

John, if you feel the rest of the image doesn't need it, just select and sharpen the main subject.

[Dan Brown]

Hi all. Dave Seeram has an action for Lab sharpening. I have purchased it (15 or 20 bucks) but I have only played with it so I don't know much about it.

Dan Brown,
Sacramento

[Joerg Rockenberger]

John, if you feel the rest of the image doesn't need it, just select and sharpen the main subject.

Fabs,

my question was more along the line what the advantages of LAB sharpening are. You seem to suggest that it avoids halos and other sharpening artifacts. In my experience, these particularly show up if I select the bird and try selective sharpening. Hence, the question if LAB sharpening avoids that and allows to use global sharpening. But I guess I was reading too much into it...

Thanks, JR

[Fabs Forns]

John, it's supposed to avoid color artifacts since it sharpens in the B&W channel.

[Joerg Rockenberger]

Thanks. JR

[Desmond Chan]

Actually, if you go too far, you will still get halos and so should adjust the Radius as necessary. In "the" LAB book by Dan Margulis, it also suggests blurring the "a" and "b" channels with Surface Blur to reduce noise (especially "b"). That's what I've been doing, blur "a" and "b" before sharpening the lightness channel. LAB sharpening is all I use now.

[Ted Scalzo]

Thanks you all this is a great find will try this soon.

[John Chardine]

Could you give an idea of how much surface blurring is appropriate here. Where is a good starting point?


Actually, if you go too far, you will still get halos and so should adjust the Radius as necessary. In "the" LAB book by Dan Margulis, it also suggests blurring the "a" and "b" channels with Surface Blur to reduce noise (especially "b"). That's what I've been doing, blur "a" and "b" before sharpening the lightness channel. LAB sharpening is all I use now.

[Desmond Chan]

Could you give an idea of how much surface blurring is appropriate here. Where is a good starting point?

Looking back, I think I've made a mistake. My apologies. I think I made it look like one should blur the a and b channel every time before the sharpening. The idea to blur the a and b channel is to hide the noise. In LAB, the noise usually shows up in b ( in the blue channel in the case of RGB and b control the blue-yellow axis). So if your photo does not have a lot of noise, I don't think you need to blur either the a or b channel. Or you can simply use other methods to reduce the noise first. As for how much to blur if you do need to, I myself cannot give you any guidance on it. Dan Margulis in his book uses surface blur to blur the a and b. So I suppose you have to know something about surface blur before you have some idea on the settings.

If you want to know more about LAB, I suggest you to get a copy of Dan Margulis' book. Mind you, his books don't seem to be easy to read. They have a lot of details though. They're advanced materials.

[John Chardine]

Many thanks Desmond.

I was taught a method to diagnose colour-cast using LAB colour and it works really well. Pure whites (whatever they are!) should average 0 on a and b channels. A colour cast will produce consistently + or - numbers on the channels. I then do a Curves on the offending channel with a goal to bringing the whites to 0, recheck and move on. The check is done in Histogram-Info, then run the cursor over parts of the image and look at the readout for each of the LAB channels.

[Desmond Chan]

Many thanks Desmond.

You're welcome, John.


I was taught a method to diagnose colour-cast using LAB colour and it works really well. Pure whites (whatever they are!) should average 0 on a and b channels. A colour cast will produce consistently + or - numbers on the channels. I then do a Curves on the offending channel with a goal to bringing the whites to 0, recheck and move on. The check is done in Histogram-Info, then run the cursor over parts of the image and look at the readout for each of the LAB channels.

Sometimes I use that, too. But I found that when I change back to RGB mode, the "corrected" white shifted a bit. By that I mean it doesn't appear to be a neutral white in RGB, i.e., not all the RGB values are the same. Not sure why though.