Not to disagree with the Moose in public, but I must...
Initially, ETTR was as per what Moose suggested, and that's to protect the
highlights. Since then, it has evolved to maximize the bit-depth available
for the converted RAW (Raw, raw) file. This is based on the theory that the
top stop of dynamic range contains half of the available bits for the
image's tonal depth. The next stop of dynamic range contains a quarter, the
next contains an eighth, and so forth. Provided that the sensor isn't
producing just noise, the bottom stop of dynamic range is represented by a
single bit and posterization is visible.
So, ETTR, if used to protect the highlights will result in the image being
underexposed. During conversion, the exposure must be brought back up and
adequate curves applied to move everything back to the right, but the
highlights will end up being compressed a bit to keep them from clipping.
(this is my prefered reason for using ETTR with the E-1).
If ETTR is used to maximize dynamic range AND tonal separation (maximum bit
availability), the image is overexposed in-camera. Inotherwords, you
overexpose the scene as much as you can to the point where you start to
clip, and then back off a teeny tiny bit.
If the entire scene is, say, a foggy morning where the entire dynamic range
of the scene is three stops, what you want to do is in-camera push the
exposure so far to the right that those three stops reside up at the right
hand of the histogram. During conversion, you "normalize" the exposure,
which means that you'll pull the exposure back down the same amount that you
pushed it up.
There are two distinct advantages to doing this with a low-dynamic range
subject. You get more bits available during capture to represent the usable
image. During conversion, we're going from a 12-14 bit capture medium, to a
16-bit per channel editing environment. As such, the problem with available
bits is not in the editor, but just in the camera. More bits mean smoother
gradients. The second reason has to do with noise. This has its roots going
back to the basis of Dolby noise reduction. The thinking is that noise is
more apparent in the shadows (due to sensor noise AND available bit depth),
so by amplifiying the incoming signal to the maximum, when we convert we
lower the signal back down to desired level, but in doing so, the noise is
also pushed down by the same amount.
As to the noise reduction, this is easily seen in my E-1 Dyamic Range
article where you can see the noise build-up in the images which were
boosted up in conversion. Meanwhile, the images which are pushed down in
conversion are squeeky clean. More or less, of course. The E-1 is a bit
unique because Kodak-Olympus applied a dithering noise to imply a smoother
tonal gradient across the entire image and to mask the shadow noise buildup.
Most Panasonic-sensor image files don't handle being boosted up as much
because the transitions in the tonal gradients are much more abrupt.
To the last point, CMOS sensors use on-chip noise-reduction. This is ALWAYS
in place and is used to balance out the noise profile of the sensor itself.
This is something that occurs directly on the sensor and has no
user-settings available. It's baked into the chip. Does it work? Yes it does
and extremely well. It's required as CMOS is normally noisier than CCD, so
this technology not only evened the playing field, but then advanced it much
further than CCD was able to go. But what happened is that CCD would add a
bit of random noise to the image, where the CMOS's noise tends to be more
fixed. Although this on-chip noise reduction does NOT smear or reduce detail
in any way because it is acting on each individual pixel, what it does do is
reduce or eliminate the random noise which gives a false texture. It also
means that in your bottom two stops of dynamic range, any gradient is
severely affected and stair-stepping occurs with reckless abandon.
(hyperbole alert--OK, under some circumstances where you are cranking the
curves around, you'll see the posterization). By using ETTR and overexposing
the scene, we are able to have more bits available for the deepest shadows.
The final presented image will be free from having any visible artifacts or
stair-stepping (posterization) in the shadows.
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/
|