Moire is when the pattern and the sensor sampling approach the Nyquist rate
combined with the fact the color pixels are not spatially on top of each other.
As such slight angle shifts, such as when you re-taped Thomas, may affect the
intensity of the color shift. Also the pixel density of different cameras are
sampling differently, not to mention the degree of filtering in the anti-alias
filter in front of the sensor. There are so many factors, I'm impressed with
and applaud your detailed investigation. A near impossible task to find nirvana.
The Siemens star image used in
https://blog.kasson.com/the-last-word/adobe-camera-raw-11-2-enhance-details/
would eliminate differences in pixel density and orientation, to some degree.
Reminds me that film, with randomness in the grain sampling, does not have this
problem of uniform sampling. The randomness of grain in film vs randomness in a
scene for digital sensors is the tradeoff.
In signal processing, the sharper the cutoff of a low pass filter, the more
there is overshoot in the step response. This is what creates the halos with
over-sharpening.
https://en.wikipedia.org/wiki/Ringing_artifacts
A Bessel and Gaussian filters, for instance, have a much smoother filter
roll-off and no overshoot. I spent many years designing filters where managers
never understood that you cannot have smooth time response with sharp filters.
If you know the subject matter is going to have moire issues, then perhaps the
best approach is to shoot taking that into account?
https://www.nikonusa.com/en/learn-and-explore/a/products-and-innovation/moir%C3%A9-false-color.html
I like Kasson's comment on ACR enhanced detail: "The good news was that it
appeared to do no harm. The bad news is that it didn't do much good."
I'm not familiar with all the terms. By HR do you mean higher resolution mode
using sensor shift technology? Seems like the best way to compensate for moire,
is that what <4,3> E-M5 II HR=>4640 is doing?
I wonder if sensor shift could be done in some sort of random way to simulate
grain?
going off the deep end
WayneS
At 2/19/2019 10:06 PM, you wrote:
>On 2/14/2019 5:05 PM, Mike Gordon via olympus wrote:
>>WSM writes: <<<<I'd like to see the Raw file for that example, to see how DxO
>>works with it. It generally seems to a better job with moire than ACR. DXO
>>has had an anti- moiré feature for a few versions. Several versions ago they
>>put priority on having the best (or close to) demosaicing algorithm. Hard to
>>tell if adobe has leapfrogged them. Would have to turn off the default amount
>>of deconvolution to tell.
>
>I'm not sure what to call this. It's not only moiré, it's false colors (as
>the below link clearly shows) and other artifacts of demosaicing color sensel
>arrays. And - it's really tricksy. Examples with one subject may not be
>representative of what will result with another.
>
>As I was looking into resolution with the GX9, I happen to have lots of test
>images. Fortunately, I have a reference standard. Shooting with HR mode on the
>E-M5 II samples each pixel location with sensels of each color, so there is no
>demosaicing to interpolate colors.
><http://www.moosemystic.net/Gallery/tech/Process/Moire_plus/Moire+.htm>
>
>The HR Raw file is quite soft. Oly deals with this by using sharpening for the
>JPEG version. They also produced an ACR/LR plug-in. Unfortunately, however, it
>uses simple USM, and generates bad halos. Fortunately, deconvolution, here in
>the form of Focus Magic, nicely improves visible details. In the second image,
>you can see clearly the crosshatch pattern of the background to Tom J.
>
>In the second row, ACR, with no sharpening or NR does a fairly nice job of
>showing the vertical part of the, background, while hinting at the horizontal,
>with moderate moire. I'm not sure whether it's default sharpening and NR
>settings improve the false color, or mask it, but it is less obvious. The HR
>image, downsized to 16 MP shows what demosaicing has lost.
>
>Third row is DxO PhotoLab*/ Not much to choose from in sharpness with straight
>conversion. In both PL cases, I added a conversion with "Lens Sharpening" set
>below the default, @ -1, as the default seemed to go too far, esp. in the
>color of his coat. With ACR default and PL -1, false color has been
>masked/corrected about the same amount. Tom seems a bit less sharp in PL, but
>the signature "Cabral" and the text "United States" below it look cleaner.
>
>The really interesting thing is what it has done with the background grid
>pattern, with various brightish white lines, horizontal dominating in some
>places, vertical in others. I think mushing out details is better than this
>creation of artificial ones. PL 0 just makes the background more so, while
>putting halos around script and text.
>
>For the 20 MP GX9 files, I'd call it for PL, slight, but definite edge in
>resolution of fine detail - if converting a file sans fine grid pattern that
>interacts strangely with PL
>
>
>Fourth row is a 16 MP file, ACR, no sharpening or NR, default settings and the
>HR file, downsampled. Even at this smaller size, it almost holds the grid
>pattern, and doesn't alter it in any way.
>
>Fifth row is the new ACR/LR "Enhanced Detail" function, which demosaics
>differently, writing the result in to an external DNG file. With no Sharpening
>and NR, it does improve the false color. With default USM and NR, it's very
>subtly different, but I can't say an improvement. I'm glad it doesn't do
>magic, as it is stone slow on my Win10 portable. Takes over the poor little
>GPU and stops everything else - just close the lid and come back later, much
>later.
>
>Sixth row is DxO PhotoLab, no Lens Sharpening, Lens Sharpening = -1 and =0,
>the default. Not much to choose from in the first two columns - except, PL
>enhances the false color in col 2, rather making it less obvious.
>
>For the 16 MP E-M5 II files, it's pretty much even, except for the false color
>in the grid with PL.
>
>Row seven is just some other 16 MP bodies, to show they all have false color,
>but it varies.
>
>As I'm interested in the best converter for my GX9 files, I've been doing some
>other testing that highlights other differences. Soon to come to a roll-over
>near you?
>
>Tedium Moose
>
>* Notes:
>The difference in image size between ACR and PL from the same Raw file are the
>result of different distortion correction. That can't be turned off in ACR/LR,
>so I left it on in PL.
>
>Why the colors are different in different ways 'tween ACR and PL on GX9 vs.
>E-M5 II is a mystery to me.
>
>Some samples are tilted because the tape I used dropped Tom, and I didn't get
>him back perfectly.
>>
>>Another good example of the new procedure:
>
>>https://blog.kasson.com/the-last-word/adobe-camera-raw-11-2-enhance-details/
>
>
>>What moiré? Mike
>
>
>--
>What if the Hokey Pokey *IS* what it's all about?
>--
>_________________________________________________________________
>Options: http://lists.thomasclausen.net/mailman/listinfo/olympus
>Archives: http://lists.thomasclausen.net/mailman/private/olympus/
>Themed Olympus Photo Exhibition: http://www.tope.nl/
--
_________________________________________________________________
Options: http://lists.thomasclausen.net/mailman/listinfo/olympus
Archives: http://lists.thomasclausen.net/mailman/private/olympus/
Themed Olympus Photo Exhibition: http://www.tope.nl/
|