>
> If you're serious on this (I assume you are), then I must start
> thinking this whole thing over.
>
Not really that big of a deal. aRGB is a better EDITING and STORAGE
colorspace than sRGB and presents a wider color gamut to an output device
which supports a wider colorspace than sRGB.
What happens here with colorspaces is that sRGB spreads the steps out evenly
across the entire dynamic range from max black to 100% lumenance of each
color channel. aRGB robs tonal steps from part of the tonal range and
applies those steps in other parts of the tonal range. When you convert an
aRGB image back to sRGB, it remaps those steps to the nearest step in the
sRGB colorspace. The selected gamma also determines the spacing of those
steps to lumenance values, so the spacings aren't equally stepped across the
spectrum.
Let's take a RAW file. For simplicity sake, let's say that the A-D
conversion is 8-bit. This gives us 256 brightness levels (steps) per color
(0-255). Unfortunately if the sensor is a linear device and the supporting
analog circuitry is not shaping the response curves prior to A-D conversion,
then one half of those brightness levels are assigned to the top stop (zone
from here on out) of dynamic range. This means that we have 128 steps
assigned to JUST one zone. But then the next zone down has 1/2 of the
remaining bits or 64 possible steps. The third stop or zone of brightness
range has 32 possible steps, the next one has 16 possible steps and so on.
By the time you get to the darkest zones, you literally have three zones
represented by seven steps. This is the "solarization" of which I speak.
To maximize image quality because of this A-D conversion, we talk about
"Expose to the Right" which pushes the maximum amount of the image as far up
as we can so we have the most number of steps in the effective range. But
this usually ends up overexposing the image. No big deal, though, because
during conversion and placement into a colorspace, we'll reassign all those
high bits down to where we really want them to be. As long as the camera
isn't doing any curves adjusting PRIOR to the A-D conversion or in-camera
prior to the RAW file writing everything should be OK. (Note the premise of
that last statement--most, if not all cameras do adjust the image prior to
digitization and will require a different ICC profile to get things to map
correctly).
When converting the RAW data into editable data, we remap all those steps to
the sRGB colorspace which pretty much, other than the gamma offset, equally
assigns steps from black to max, or to aRGB which puts more steps in part of
the tonal range than others. By stealing bits from, say, the middle tones,
we are able to assign more bits to the highest and lowest values.
While we are at it, we also have the choice of 16-bits (two bytes) per color
or 8-bits (one byte) per color when we convert and save. Working in 16-bit
mode makes editing a whole lot easier because you don't get solarization
artifacts when making major adjustments.
When we convert between colorspaces, we have the choice of preserving
colors, saturation or high/low thresholds, but we can't save them all, so we
decide based on personal experience, preference, the image itself and output
medium which one of these traits we wish to preserve.
I mentioned a day or two ago how we can end up with part of the image being
represented by only a few bits of data. Let's assume our original image was
four stops under exposed. That means that the entire image dynamic range
could be represented by no more than 32 possible tonal steps!!! During
conversion and editing we pull up the exposure which now spreads those 32
original steps across 256 steps (8-bit) or 65536 steps (16-bit). Just
because we spread those 32 original steps out to 65536 steps doesn't mean
that it has 65536 steps worth of tonal values--it still is just the original
32 steps with a whole lot of gaps. Now, let's apply some additional curves,
color and constrast adjustments to this image file. Every time you do this
you are pushing values up or down and potentially eliminating original
values. This is why we can end up with a portion of the image getting only a
handful of original steps preserved through the image editing process. Of my
original 32 steps, if the dynamic range of the picture is four stops, and
the steps are equally spaced, then each stop has only 8 steps. Depending on
where in the dynamic range we're looking, how the colors are mapped, the
original A-D conversion, etc., etc., you can end up with far fewer than 8
steps per stop or zone. That's 3-bits.
Before anybody thinks that we don't frequently do 4-stops of adjustments, I
challenge this by a simple test. Import a RAW image file with no adjustments
other than exposure offset. Keep the colors straight and don't apply any
curves adjustment as the ONLY adjustments made prior to the assignment to
colorspace is exposure compensation and highlight recovery. (Note, there may
be some variances on this from one converter to another, but generally
speaking all curves and color adjustments are made after the RAW file has
been remapped into a 16-bit working colorspace). The file should only have a
max black and a max white if the picture is of a high contrast scene. But
the colors should be pretty flat. Work in 8-bit mode for this, as it's
easier to see the effect. Use your "inspector" and look at the RGB values
for given pixels on the image. Now, start editing the picture, changing the
curves, saturation, etc. Now use your inspector to look at the RGB values at
those exact same pixels again. It is NOT unusual for you to have adjusted
parts of the image between two and four stops. The overall brightness of a
spot may not look different to the human eye, but the RGB values will
reflect a change. If the values change from "32,64,128" to "32,256,128" the
green channel for that pixel has increased two stops. Overall, the image may
not have changed drastically, but each pixel may have. Usually, this just
shows itself as an increase in noise, but on an expanse of a single color,
it will show up as solarization.
Too much information?
AG
--
_________________________________________________________________
Options: http://lists.thomasclausen.net/mailman/listinfo/olympus
Archives: http://lists.thomasclausen.net/mailman/private/olympus/
Themed Olympus Photo Exhibition: http://www.tope.nl/
|