Good pointer Dave. I use Pixel Genius's tools for all my sharpening these days (referred to towards the bottom of that article).
For the technical background of why sharpening the L channel in Lab is better than sharpening in RGB, I'll have a go at explaining it:
Sharpening is done by enhancing the contrast at edges. Thus at edges we make the dark side darker and the light side lighter. Do it too much and you notice "halos", but the issues go deeper than that.
In an RGB image, the sharpening IS APPLIED TO EACH CHANNEL INDEPENDENTLY. Thus at an edge in red we'll make the lighter side a bit redder and the darker side a bit less red (and so on for green and blue). Unfortunately this can cause colour shifts at edges. Consider this (agressively sharpened, and then enlarged 500%) edge between yellow and grey:
The yellow side of the edge hasn't got any lighter (it can't) but the grey side has become blue! That's because the reds and greens have been darkened.
In the L*ab colour space, instead of red/green/blue coordinates we use lightness, and "a" and "b" which together define the colour. By sharpening just the L channel we get this:
However, note that the Lab colour space is huge: it describes lots of colours that simply don't exist (if that makes sense). At least, it describes a colour space that's MUCH larger than your RGB working space (e.g. sRGB, AdobeRGB, etc). And in each of these channels, in 8-bit
mode there're only 256 steps available (and many of those usually impossible colours). This means that
in 8-bit mode Lab is often less precise than RGB! The same thing happens with use RGB spaces such as ProPhoto RGB. Not only can posterisation be introduced, but the precision in some calculations is reduced.
So
I never use Lab unless I'm in 16-bit mode. The above L-sharpening was done by converting to 16-bit before transforming to Lab, and converting back once the image was in RGB again. Have a look at what happened when I left it in 8-bit
mode all the way through:
Where did that texture in the grey come from? From rounding errors in the colourspace transformation on the way to Lab and then on the way back again! The 16-bit version is perfectly smooth.
So there's an example of why sharpening the lightness channel is theoretically better. This is only a small part of the picture (pun not intended) and is often very subtle, but should provide some insight into why there are so many different ways of sharpening (and the comprehensive ones often use L sharpening, even if they don't make it obvious to you the user).
Posted: Wed Nov 16, 2005 6:24 pm