Dpreview's ACR myth distorts their reviews; why do this?

Started Nov 18, 2008 | Discussions thread
ForumParentFirstPrevious
Flat view
Tom Christiansen Senior Member • Posts: 2,239
Dpreview's ACR myth distorts their reviews; why do this?

In the "Editorial Blog" at http://blog.dpreview.com/editorial/2008/11/adobe-camera-ra.html

Lars Rehm wrote on November 04, 2008 that:

Secondly, especially when analyzing image noise, we want to show a
camera's image output in its 'purest' form, i.e. as captured by the
sensor and as little as possible altered by sharpening algorithms,
noise reduction or any other forms of image processing. Again, ACR
does a pretty good job in this area. To demonstrate this I have
included example crops from two images which both have been processed
from an ISO 3200 Canon EOS 50D RAW file, one in ACR 4.6, the other in
Canon's Digital Photo Pro 3.5. Sharpening and noise reduction were set
to zero in both converters, then we applied an identical unsharp mask
to both images, so in theory we would expect pretty similar results.

Reality looks slightly different though. The ACR image is less
sharpened and shows visibly larger amounts of chroma noise. That tells
us two things. Firstly DPP is doing quite a good job at noise
reduction and is probably the better choice for cleaning high ISO
pictures of the 50D (ACR's NR can't match DPP's even if you turn it
up). However, it also means that DPP applies at least some chroma
noise reduction (and sharpening) even when NR is set to zero which
renders the software pretty much less useless for our purposes.

This is simply false. Dpreview are propagating a myth that does a
disservice to your whole site's credibility. You need to rethink,
recant, and re-evaluate how you're doing these conversions.

Here's why.

Just because one converter gets better results than another when
sharpening and NR are set to "0" DOES   NOT   MEAN that the better
converter is secretly applying sharpening and NR behind your back!

All software, all algorithms are not equal. Demosaicking is an art, and
a developing one at that.

Only the maker of the chip (and if different, also the company buying
that chip) knows the precise spectral performance characteristics,
including metameric effects which vary depending on white balance and
ambient lighting, of the sensor's different CFA dyes.

This information is not in the raw file; that's why ACR has to guess,
and the manufacturers do not. Adobe does not know . They guess. They
test. They guess again. But they still do not know .

Even slight errors, perhaps misunderstanding the sensor response under
various lighting, are magnified in a low signal-to-noise situation. A
metameric miscall could show up as noise in ACR, noise the manufacturer
knows better than to introduce.

That's why, when Nikon's D3 and D300 debuted, ACR was abysmal at
demosaicking the NEF files. And that you call that a "level playing
field"?! It isn't. It is as slanted as can be. While Adobe later issued
improved profiles, they are still only guessing.

Canon and Nikon know a great deal more about how the chips and their
cameras behave than Adobe ever can. They know the dyes', they know what
white balance does to metameric tricks with those dyes, they understand
the signal processing that occurs in hardware due to exposure-index/ISO
tweaking, and they know how to weight that signal due to fill factor and
microlenses (such as the double-layer of microlenses on the D3).

The manufacturers, unlike Adobe, can react to these and many other
factors by varying the algorithms used when demosaicking. This can
trivially lead to lower noise and improved sharpness without the
application of any "noise reduction" or "sharpening" going on behind
your back.

A simple demo of this can be seen in how Nikon, an optics company,
understands the math behind the light-scatter producing lateral
chromatic aberration in the periphery of many lenses, especially wides.
They can take this into account when demosaicking to produce a notably
cleaner and sharper image. Look at power lines in the frame edges to see
what I mean. They resolve into better per-pixel sharpness when they
realign the spectral frequencies back where they should be. It's like
apochromatics in math instead of optics. That's not secret sharpening!
That's better math.

I have no doubt that other clever elements are factored in that we
haven't even thought of. For example, if the white balance is 1700K,
should you really be weighting the "blue pixels" the same way you do
when it's at 5400K, or the "red pixels" when at 15000K? If you pay less
attention to the noisy channels during the demosaicking process, you'll
again get a cleaner image with less noise.

Your entire premise that ACR gives a level playing field is
demonstrably false. And I am hardly the first to point this out. If
your premise were true, D3 and D300 NEFs would look as good under ACR as
under the Nikon converter, BUT THEY DO NOT ! Similarly, if it were
true, then the Canon raw files would look as good under ACR as under
Canon's converters, but as you yourself note, they do not. Therefore,
your premise is false.

But that thought, that you might be wrong and why, does not seem to
enter your mind. Instead, you jump to an erroneous conclusion: that
somehow, somewhere, some "secret" sharpening or NR is going on.

A far simpler explanation is that the manufacturers better understand
how their cameras behave. And this has been repeatedly demonstrated.

Yet still you pretend ACR gives a level playing field, while nothing
could be further from the the truth!

Why are you continuing to spread this myth? Is it truly because you
simply do not understand the matter? Or do you have other reasons to
pretend your myth to be true?

I can certainly think of a few, but I'll not impugn your reputation with
hypotheticals until you explain why you are spreading falsehoods about
ACR and level playing fields.

-- hide signature --

tom

ForumParentFirstPrevious
Flat view
Post (hide subjects) Posted by
tko
ForumParentFirstPrevious
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow