Telephotography of restricted areas
Color Correction
You are on your own here. A good guideline for color correction is to
have the right edge of the histograms line up. Vuescan will do it's own
white balance by right clicking on where you think it should be
white.
Two obvious defects can be seen in this image. [Well, three is you
think it should be sharper, to which I say, move the base closer to
Tikaboo, and prevent forest fires in California.] The image
darkens at the edges, which is to be expected. As the contrast is
cranked up, the darkening at the edges becomes more pronounced. The
second problem is that the color cast changes as you view the image
from top to bottom. This will not be as big of an issue after the image
is cropped. Both effects can be compensated with gradients in
photoshop. The color banding is due to the fact that different parts of
the photograph have passed through different amounts of
atmosphere.
Not really a defect, but the white of the roof is really out of kilter
with the rest of the image. This is easy to adjust in photoshop. First
read the image into photoshop in LAB format. [The operations change
from revision to revision of Photoshop. Version 6 is used in this
example.]
Select ->color range ->highlights
In the case above, the mask on the hangar looked like this at first:
This making would cause any tweaking to look a bit rippled. The best
thing to do is simplify the mask Borrowing a trick from
intergrated circuit mask processing, expand the mask, which causes the
slivers in the mask to merge, then contract the mask by the same
amount. In this example, the mask was expanded by 5 pixels, then
contracted by 5 pixels.
Select->modify->expand using a value of 5, then select->modify->contract using the value of 5.
To make the masking operation smoother, feather it by a small amount,
in this case 4 pixels. Then dim the bright spots using the brightness
adjustment with a value of -5.
select->feather with a value of 4
image->adjust->brightness with a value of -5
select->deselect
The effect is very subtle.
Now to reduce the blue band in the middle. It can be adjusted with
gradient, but you will first need to set up the foreground and
background colors. I set the foreground to 90 and the background 166.
The exact numbers are not critical, though the average of the two
numbers should be 128.
Convert the image back to
RGB. Select the blue channel. Copy it, then create a new image with the
blue channel. Create a new layer. Convert this layer to "hard
light." Slide the opacity control until the image looks balanced.
image->mode->RGB
channels->blue
select->all
edit->copy
file->new->ok
edit->paste (This is in the new image. You need the
old image to past the processed blue layer back into, so don't alter
it. The pasted layer will be layer 1 in the new image.)
layer->new layer->ok (This will be layer 2)
Select the gradient tool
Select the horizontal gradient
To create the gradient, place the mouse at the center of the image.
Holding the left button down, move the cursor to the top of the image.
The gradient will now be applied. You ned to adjust the opacity to
suit. This is subjective of course. Sometimes it helps not to be too
close to the screen when making the adjustment. Once you like the
image, flatten it to get rid of the gradient layer
layer->flatten image
Here are the results from the horizontal gradient application.
Now adjust the darkening at the edges with a radial mask. You need to select the radial gradient
The step are identical to the horizontal gradient adjustment.
Here are the results from the radial gradient adjustment.
The green channel requires a radial gradient adjustment.
Below is the "before and after" for the field flattening operation.
There is still a color cast, but the overall color is smoother across
the image. This may be overkill for one photograph,but the field
flattening techniques are needed for producing panoramic displays, i.e.
pasting together different photographs. Otherwise the images will have
a very visible hot-spot for each photographed used.
At this point, the blue channel is still the 800 lb gorilla in the room.
The blue channel doesn't have much detail, and it has the most
noise. At this point it helps to understand human vision, which is a
combination of hardware (eye) and signal processing (brain). For
simplicity, I'll just refer to the eye, but remember the brain is part
of the equation. The eye
sees detail in black and white, and then paints it with color. [OK, a
gross simplification.] You can remove some of the blue channel noise by
converting the RGB image to LAB, replacing the L channel with a new
image comprised of just the R and G channel from the original RGB
image. Some of the noise from the blue channel will find its way into
the AB (color difference channels), but the luminance (L channel) won't
have the blue noise. This is cheating slightly on the color of the
image, but you will never get perfect color anyway, so you might as
well produce a sharper, cleaner image.
Below is the new blue channel produced by this process.
The final result is below. The grain isn't that much of an issue since the image size is reduced.
