Noise Reduction / Smart Sharpen
Let's have a look at what we're talking about and then run an example.
First click on the two thumbnail images, save the full size jpg's and
have a look at them at 3x (300% zoom). These are captures from a digital
video which was running in poor light - so the starting image quality
is terrible. The background and the bare skin of the legs is a good place
to look. What you will see is that, aside from the improvements in color,
the processed image has lost most of that terrible mottling while at the
same time the overall image seems to have sharpened moderately. Nothing
comes without some cost, so there is still some pixellation in the areas
of sharpened detail, most noticeable around the edges of the face and
hands. But still overall it's a pretty remarkable improvement at normal
100% zoom - and one of the key reasons for this is that image noise is
far more noticeable in flat areas of color than it is in detail areas.
You can delete the images, they are not used further here.
Doing it
Load "Set 4noisesharpen.atn" into the actions palette as outlined
in Task 1.1, and run it on "ws-photo.jpg". You can compare the
original Set4a output here
with the noise reduction output here.
What we get with this image is not nearly as impressive as the samples
on the right. There are two reasons. First, the starting image was in
far better shape than the video captures, and second the sharpening applied
in the process is much less than the unsharp mask sharpening that was
applied in the original Set 4a actions.
Lets try again with an additional unsharp mask added at the end of the
noise reduction. Load "Set 4noisesharpen2.atn" and run it. The
new output image is here.
This is sharper and different from the original Set 4a output but not
necessarily a lot better. What you can see however, if you look in the
model's shadow on the wall, is that the mottling in this shady area has
been removed. The same is true if you look with 200% zoom at the models
skin. It may be that the amount of sharpening is now too much and should
be lessened.
Enough of that - getting something like this right can involve quite
a lot of experimentation, however, as the thumbnailed samples above show,
in certain situations the work can be well worth it - especially where
you have a large number of images shot in the same lighting conditions.
What is Happening
The basics of the process are
- applying a series of masks, often separately by color channel, which
then allow one to get rid of that mottling noise pattern,
- applying this noise reduction in areas of flat color,
- avoiding applying the noise reduction, which involves Gaussian blurring,
to detail areas where it is likely to damage the image,
- then reversing the technique to sharpen edges while not sharpening
flat color areas.
The reason I included this brief final section is twofold. First it gives
some indication of how far you can go with the automation processes of
Photoshop Actions. Second, I see working my way through a set of actions
like these as being one of the most powerful cheap lessons available for
learning how to use the full potential of Photoshop image manipulation.
Further Sources of Info
Fred Miranda : home and "actions"
here
The noise reduction I have used above is a jigged around version of Fred's
Nikon Coolpix 5000 actions which do cost you a few dollars but which open
a whole world of image manipulation. I should add that my application
of it here does no justice to the quality of the product.
Luminous Landscape: John Brownlow on Smart Sharpening here
(with free actions download) - this is a superb introduction to the technique.
|
By Forest
Part 1 | Part 2 | Part 3
