Images with large amounts of colour noise can lead to bigger files when
using Jpeg compression, Matthew Page
describes his technique for reducing the noise, without sacrificing
detail. Photos by Nic Duncan
There are times when splitting an image into
its component parts is
extremely useful. Hopefully through this little tutorial you will get
an appreciation for layers, layer blend modes, and luminosity and
First of all, a quick primer about image components. You can represent
an image in a number of ways. You are probably already aware that you
can describe the image pixels in Red, Green, and Blue (RGB) quatities.
However, there are a number of other ways to represent the image
information. The JPG picture format actually splits the image into 2
separate components – luminosity and colour.
The Luminosity is, if you like, the grey scale part of the image. (My
tutorial on “Colour Punch” described how to get
black and white images easily. This is the luminosity content).
The Colour part contains hue and saturation together, and can look very
strange on its own. However, being able to manipulate the colour part
can be extremely helpful in reducing colour noise in an image.
Colour Noise and JPG
The JPG format records the grey scale part almost as-is. The colour
part, though, is run through a complex mathematical process where the
high frequency information is thrown away (hoping you don’t
notice it!) so the image size can be reduced.
When an image contains either a lot of colour information, or a lot of
colour noise, squishing down the JPG file can be very difficult
you find you cannot reduce the file to a specific size (e.g. 200k!)
without losing critical image information.
To help the JPG encoding along, we can tackle the colour information
manually and make the encoding process easier – thereby
some file space (particularly important where file size restrictions
are in place on certain websites!).
A Sample Image
|Here’s an image (thanks
that contains a great deal of colour noise:
|And here’s a 200% view:
You can see the noise in the skin tones.
Here’s the process to split out the colour component of the
|First, add a new blank layer
|Now fill the layer with 50% grey
(Edit / Fill)
|Now change the layer blend mode to
You should now see the colour information… but we need to
“preserve” that information in a layer, so:
|Hold down the ALT key and select
the menu “Layer / Merge Visible”
(Alt-Ctrl-Shift-E). A new layer will appear (I’ve named it
component” so you can see what’s going on.
Before we ditch the grey layer, we have one more job for it.
|Hide the “Colour
component” layer (by clicking on the little eye in the layers
Change the blend mode for Layer1 to “Colour”
Now you should be able to see the black and white image information.
|Hold down the ALT key and select
from the menu “Layer / Merge Visible”
(again I’ve renamed the layer so you can see what’s
You can now delete the “Layer 1” layer as it has
done its work.
Now, that’s done the image split. We have a separated colour
component, and a separated luminosity component. It’s time to
filter the colour component.
Filtering the Colour
For this little trick I’m going to take you into an
world of other filters! There is a very useful filter called “High
that keeps the detail in an image and throws away the non-detail. You
may have heard of “High Pass Sharpening”
as a general
technique, but I’m going to show you “High
” as an alternative.
Putting it back together
|Duplicate the “Colour
component” layer (select the top layer and press Ctrl-J is a
Now call up the menu: “Filter / Other / High Pass”
|Here you need to pick a radius that
just shows the details you want to
get rid of. For this 1500x1000 image I chose 2.5 – it will be
dependent. If you go too large on the radius you will throw away too
much colour information and colours will bleed in the final stages.
You should see a
mainly grey image with feint patches of colour detail.
|Now change the blend mode to
“Overlay” (I’ve renamed the layer to
“Denoise” to help)
Woah! More noise! Loads more in fact. We have emphasised the noise
(this is high pass sharpening) but...
Now invert the “Denoise” layer (Ctrl-I) and
voilá! The noise disappears!
|The final thing to do is to group
the “Denoise” layer so it only
affects the “Colour Component” layer. Press
Ctrl-Alt-G (or the menu
“Layer / Create Clipping Mask” – in CS2)
So, at this stage we have a Luminosity component, and a filtered colour
To re-apply the colour information to the luminosity and glue it all
|Select the “Colour
Change the layer blend mode to “Colour” (from
And that’s the image back!
If you can see a difference, then you’ve gone too far.
when you save for web you should find a reasonable difference between
the compression ratio you need to achieve a fixed JPG file size.
For my last trick
|Here’s that 200% closeup
Less skin noise for sure, and the save for web could achieve a higher
quality for the same filesize.
Since we have split the colour information we can also adjust
saturation in a non-destructive way.
|By adding a grouped-layer set to
“Saturation” blend mode:
Now, pick a soft brush set to 10% opacity and:
- Paint black to reduce saturation
- Paint bright red to increase saturation
You can, of course turn off that layer to compare. And, you could also
re-colour the original image by painting on a layer set to
The same principles apply to modifying the Luminosity component
can make changes without affecting the colour (hue &
Aren’t layers wonderful?!