| Goodbye Wavelets
High Pass Filter Processing for Lunar Image Enhancement
|
| Cheap video based digital imagers have
changed the face of lunar and planetary imaging in recent years. Anyone involved in that revolution has used wavelet sharpening as the most common
method of resolution enhancement. While wavelets do get the job done with a
high degree of flexibility and power there has always been a price to pay in
artifact generation, noise and destruction of tonal balance. Here's a
typical example: |
  |
| This comparison shows
everything that's good and bad about wavelets. They do a great job of
bringing up the detail in the image but they have also can bring up a lot of
noise. Also note how some of the bright crater rims have picked up an edge
artifact and and have been pushed into clipping. There's also a general loss
of tonal gradation and loss of mid-tones giving the processed image a very
hard look. When working with wavelets a good part of your workflow is
concerned with minimizing or correcting these negative effects. In most
cases your ability to enhance the image is very much limited by your ability
to deal with these effects. Noise can be reduced or eliminated but at the
cost of the finer detail in the image. Tonal quality can be
recovered to a certain extent by careful use of levels and curves but it's
constant battle. The good news is, there is a better way. |
| Enter multi-pass high pass filtering.
Found in various versions of Photoshop and a few other programs including
the widely available Photoshop Elements, high pass
filtering offers a way to enhance the fine details of an image with far
fewer negative effects then traditional wavelet processing. Basically, all a high pass filter does
is suppress the
low frequency detail in an image. In audio this would be like turning down your bass control while turning up the treble. When you do the same thing to an
image, here's the result:
 |
| This ghostly image consists
largely of areas of high contrast or fine detail in the image. While
of little use on it's own this image can be used as a mask to emphasize the
parts of the image where contrast (detail) is present. Here's how it's done
with a full frame image in Photoshop Elements: |
| |
 |
| This is the starting point,
an aligned, stacked raw image. deconvolution sharpening has already been
applied but nothing else has been done to improve the image. The next step
on the processing road is to enhance the detail and overall sharpness of the
image. To do this with high pass filtering you start by making a duplicate
layer. To that layer apply the high pass filter. In Photoshop Elements it's found
under "Other" in
the "Filter" menu. Adjust the Filter Radius setting so that most of the
content of the original image is visible as we want this first pass to be
rather heavy. In this case as value of 7.1 gives a result that's about
right, it looks like
this: 
|
| The next step is to combine
this layer with the layer that contains the original image. Elements gives
you a number of choices here but for this step in the process change the
selection in
the layer control box from "Normal" to "Soft Light".
This will cause the two images will be
combined in such as way as to emphasize the detail passed by the filter:
 |
| The original image is on the
left, HP pass filter layer applied is on the right. The filter provided a
useful increase in image sharpness without any of the noise and artifacts of
other methods. If the effect is too strong simply use the opacity slider to
adjust the strength of the effect. Once you're satisfied with the result go
to the layer menu and flatten the image. |
|
| One of the really useful
aspects of this approach is that unlike other sharpening methods you can
make multiple passes using different combinations of methods and
opacities. My standard method would be to take the result of the first
pass and repeat the process two more times. For the 2nd pass I'll generally use a smaller
pixel radius value then the first. This keeps the effect under control
and selectively targets the smaller details in the image. In this case I'm
stepping down to a pixel radius value of 3.5
 |
| This is the result of the
2nd "Soft Light" combination at a filter radius 3.5 with an opacity
setting of 80%.
Getting this level of enhancement with wavelets would have taken a fairly
aggressive setting with it's attendant high noise level. Here the background
tones are still extremely clean. That's good news because it opens the
possibility of processing the image further with a 3rd pass.
For the last pass select an even smaller filter radius
then you did in pass 2. In this case I'll use a value of just 1.5 pixels,
targeting the very smallest details in the image. For this last pass to
really be effective you'll need to use a more aggressive combination method. With
most images the most effective choice for the 3rd pass is "Vivid Light".
As before, adjust opacity to balance the effect. You don't want this to be
too heavy, you're looking for a slight and controlled improvement.
Here's the result with the 3rd pass applied compared
side by side with the original raw image:
|
| The result as you can see is
extremely
good. Close examination of the image shows very little noise enhancement and
a higher level of detail then can be achieved with wavelets after the
required noise reduction and adjustments. |
| |
| |
| Another handy
application of the technique is to also use it as a Levels adjustment layer
that won't cause the heavy gaps and attendant posterization that can happen
when pushing for a large Levels adjustment to the completed image. To do
that simply apply the Levels control during any HP sharpening pass after
selection the combination method but before flattening the layers. Use the
midpoint control to adjust the brightness and then flatten as usual. Here's
an example: 
|
|
As you can see its a rather effective technique. |
| Here are a few more
general HP filtering examples: |
| |
  |
| Because of the extra "push"
they often need on post processing sharpening, high sun angle images are
usually the most difficult to control in terms of noise. Some noise
reduction was required as a final step in this image however the strength of
the reduction required was very low. |
| |
  |
| And finally a low sun angle
example. Noise reduction was not required.
As these examples show, this is a very powerful and flexible technique that deserves a
spot in every high resolution imager's bag of tricks.
Tony Gondola, April 2005
|