In the previous steps
in processing we have kept the scientific information in the image
essentially intact. The files that were written out to the RAW format
are deep 16 bit files and captures something of the dynamic range of the
object. However these images are not very satisfactory to view.
There is information hidden in the bottom few percent range of the image.
For example, in the image of M83, the galaxy arms have brightness values
around 4300, against background levels of 3700. The maximum value in
the image is 19300, so there is much interesting information in the bottom
4% of the image.
The process described below aims to transform the dynamic
range of the image so that details, both in the galaxy arms and the bright
region around the core are visible. This is known as contrast or
histogram stretching. There are several ways of doing this. The
log transform in Iris is one way - that has the advantage of being
repeatable and can be consistently applied to different images, or different
colour channels. However this has the disadvantage of applying a
drastic stretch, particularly to the fainter parts of the image and tends to
emphasise low level noise.
I use the adjustment of Curves in Photoshop to do
histogram stretching. This lets me control better the contrast of the
image, the noise, and lets me stop bright stars bloating. However the
process is not very repeatable. In order to maintain the relationship
between colour channels as far as possible throughout the stretching
process, these transforms should be done on all these as the same time,
which is the reason for reading in the combined 16 bit RGB image.
|Reading the images into Photoshop
First open the Y image. In Photoshop do File, Open. The Raw
Options box opens. Note that the dimensions of the image has to be
specified. The Y image has only 1 channel and the file is 16 bits with
IBM PC byte order. When the file was exported from Iris, we specified
that the header size would be 0.
Here's what the image y.raw looks like when first opened. Note that
it is quite dark and that it's flipped vertically. Do Image - Rotate
Canvas - Flip canvas vertical.
The image histogram shows why it is so dark. The values are bunched
up towards the bottom of the range. At this point I used to
slide the white point slider (the right one) to the left to the highest tone
values, but I have since found that this is unnecessary and in fact produces
washed out highlights.
To do histogram stretching, do Image
- Adjustments - Curves... The curve will be straight to start with,
but grab a point on the lower left of the line and pull it up. Let's
call this a Positive Curve Stretch. The
fainter parts of the image will start to appear.
The same curve adjustment is repeated.
More of the arms now appear. The histogram now shows better use of the
dynamic range with the bright stars approaching saturation. Any more
stretching and these stars will start to bloat. The next step aims to
Preventing bright star bloat
Now select the brightest stars by drawing, with the
Elliptical Marquee Tool, small
circles around them. Make sure that Feather is set to 2 pixels.
Do Select - Inverse. See the image below left. Now all further
operations will not apply to these stars. Another curve stretch
results in the image on the right. The background is now too bright,
and there are darl halos around the selected stars because their development
has been held back.
A look at the histogram shows that the high end of the dynamic range
is well used, but the low end is too bright. Now the background of the
image needs to be reduced in level. This could be done by simply
grabbing the black point slider and moving it to the right to where the
graph starts to climb. However this tends to emphasise low level
noise. I prefer to do the reverse of the curve manipulations before to
quieten the background.
Quieten the background
The following curve,
which I will call a Negative Curve Stretch, done twice over,
gives this new image. The fainter parts of the galaxy are now visible
while the details are maintained in the highlights.
The histogram confirms a good use of the dynamic range.
Compare this with the previous histograms. The spikes in the low part
of the dynamic range have been reduced. The bright parts of the image
now extend all the way to the top of the dynamic range, without any
clipping, which would have been indicated by a spike at the very right of
the histogram. However the background is still too light, and the
histogram should probably be stretched some more.
Negative, Positive, Negative Curve Stretches in that
order gives the following image and histogram.
This is now close to the desired result. Notice
that the histogram is still smooth, after all that manipulation! This
would not have been possible with an 8 bit image.
The next step is to match the bright stars back to the
rest of the image. In the image above they have bright halos, which
will need to be corrected. As the previous transforms acted on the
whole image except the stars, I now invert the selection, Select - Inverse.
Then Feather the selection to 2 pixels to smoothen the effect. Here's
a closer view of the stars in the image above, the subsequent Curve, and the
result (de-selected for clarity).
The stars are now smaller but there is still a bit of a
bright halo. At this point I examine other selected stars that might
have a brighter background, such as those embedded in the galaxy arms, and
de-select them (using the Lasso tool set to "Subtract from Selection") if
they look fine against the background. For the stars with the darker
background, some further adjustment is necessary. The final
Curve transform on the stars and the result are shown below. Look at
that bright star near the top right corner. It was not selected
because I expected it to be cropped out in the final image. Notice how
large it is. The other bright stars would have looked like that if
they had not been held back. The stars are still not perfect,
but a slight halo actually helps to show star colours, and there will be
opportunities to do further adjustments in the next stages of processing.
Unsharp masking may be done at this stage to enhance detail.
(Part 1, Introduction)
(Part 2, Aligning and Stacking in Iris)