Pages: 1 [2]   Go Down

Author Topic: Fake detail, on a feather  (Read 10014 times)

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
Re: Fake detail, on a feather
« Reply #20 on: July 15, 2013, 03:01:12 am »

Hi,

LR 5, but I also made some quick and dirty checks with Capture One and RawDeveloper. The results are very similar. The amount of color artifacts differ but monochrome aliasing, what I was looking at, is similar.

Did not really try to supress color moiré. Regarding color artifacts on LensAlign target there was a lot with both LR
5 and C1. Raw developer was not imune either. I decided to skip Capture One Pro, I own it, but we make no friends. I use it for test, but not for creative work.

Best regards
Erik

What software did you use for processing Erik?
Logged
Erik Kaffehr
 

32BT

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3095
    • Pictures
Re: Fake detail, on a feather
« Reply #21 on: July 15, 2013, 03:26:03 am »

Therefore, only different aliasing amounts can cause these false color issues,

Exactly, and different aliasing amounts are caused by different position, not by different density, since there is no difference in density. That's why a merrill sensor doesn't exhibit false color issues. (lol: at least not because of sampling position! ;-) )

Luminance aliasing can not always be removed easily because it involves multiple different spatial frequencies, folded back from higher spatial frequencies from beyond Nyquist and they show up as multiple lower spatial frequency aliases. It requires elaborate reconstruction, by ...

No, aliasing can always be removed very easily by (excessive) blurring. That's irrespective of the fact that you apparently do not want blur. IF you do not want blur, THEN you may resort to all kinds of resolution-preservation and -reconstruction techniques, but it should be clear that that is exactly what is happening: trying to reconstruct an image from incomplete information, because the correlation within the incomplete information allows for some apparent resolution increase. Even up to the point where the aliasing artefacts themselves may contribute to the perceived structure.

And like you mentioned, these artefacts should preferably contribute to the existing patterns. However, these artefacts are also directly related to texture reconstruction, and it is especially the latter which photographers are sensitive about imo, either consciously or subconsciously. That is why I believe there is somewhat of a disconnect between many of the forum members here where some like to scrutinise detail artefacts as structure reconstruction, and some others are then telling us that they simply don't like the "look" of a converter.



Logged
Regards,
~ O ~
If you can stomach it: pictures

yaya

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1254
    • http://yayapro.com
Re: Fake detail, on a feather
« Reply #22 on: July 15, 2013, 07:55:37 am »

Hi,

LR 5, but I also made some quick and dirty checks with Capture One and RawDeveloper. The results are very similar. The amount of color artifacts differ but monochrome aliasing, what I was looking at, is similar.

Did not really try to supress color moiré. Regarding color artifacts on LensAlign target there was a lot with both LR
5 and C1. Raw developer was not imune either. I decided to skip Capture One Pro, I own it, but we make no friends. I use it for test, but not for creative work.

Best regards
Erik


Was that Capture One 6 or 7?
Logged
Yair Shahar | Product Manager | Phase One - Cultural Heritage
e: ysh@phaseone.com |

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8913
Re: Fake detail, on a feather
« Reply #23 on: July 15, 2013, 08:18:29 am »

Hi Oscar,

If Erik doesn't mind, I think this exchange about the root cause of visible aliasing artifacts may add some insight, therefore I've prepared some examples (attached).

Aliased information cannot be distinguished from real information because both are recorded together in the same sensel position. Therefore, only different aliasing amounts can cause these false color issues, and all that the demosaicing algorithms can do is iterative reduction of the local color differences where RB and G channel luminances significantly differ.

Exactly, and different aliasing amounts are caused by different position, not by different density, since there is no difference in density.

If position would change the amplitude (not the shape of aliasing because we sample a different position in the object space), then it should be reflected in the difference between the 2 Green filtered sensels that are only 1.4 diagonal sensel-pitch widths apart.

Have a look at the first attachment. I've taken a crop from the Raw sensel data, without demosaicing, and extracted the 2 Green channel data sets, and removed the Red and Blue and the other Green sensel data. So each image is essentially an almost point sample of 1 CFA color channel with other channels eliminated.

The amplitude of the Green channel aliasing shown inside the yellow Nyquist limit circle, is basically identical. There is also a difference in position, which is caused by sampling a different position in object space. Maybe that is what you were thinking of. But that Phase shift will be dealt with by the signal reconstruction process AKA demosaicing. Also the seemingly aliased region outside the Nyquist limit circle will be reconstructed to a smoother approximation of the non-discrete original scene content. So the choice of Raw converter also plays a role in all this.

Now, compare it with the Red channel aliasing amplitude inside the yellow circle of the second attachment. A combination of lower (2.8x sensel-pitch) sampling density and diffraction of this f/5.6 shot significantly reduced the amplitude of the Red channel aliasing. The Phase shift is again different from all other B/G1/G2 sampling positions, therefore the non-aliased signal outside the yellow circle also aligns differently, but that reconstruction is the task for the demosaicing algorithm. The Blue channel, while also sampled with the same large sampling distanceas the Red channel, is much less affected by diffraction and therefore shows more amplitude of its aliasing inside the yellow circle.

As is hopefully clear, the amplitudes of aliasing (inside the yellow circles) differ quite a bit. That their Phase also differs is because they are sampling different parts of the scene, with a 1 sensel offset. That Phase difference will be used to reconstruct the luminance portion of the image and, as shown earlier, quite successful because some 93.6% of full luminance resolution can be restored. The Chroma portion of the image, which in normal (not stress-test test-chart) images has a much lower spatial frequency in the actual scene, can therefore be reconstructed with relatively good fidelity to our eyes, except for where the aliasing differences throw a spanner in the reconstruction works.

So to summarize, the majority of the effect from sensel position is related to sampling different scene content which will be reconstructed as different luminance detail. The aliasing amplitude differences are what will cause reconstruction artifacts, because they are combined with actual phase shifted luminance detail and cannot be separated afterwards. Afterall, one cannot unscramble an egg omelet.

Hope that helps to clarify some of the causality.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
Re: Fake detail, on a feather
« Reply #24 on: July 15, 2013, 08:25:13 am »

Hi,

7.1.3, I think.

Best regards
Erik


Was that Capture One 6 or 7?
Logged
Erik Kaffehr
 

32BT

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3095
    • Pictures
Re: Fake detail, on a feather
« Reply #25 on: July 15, 2013, 09:22:42 am »

If Erik doesn't mind, I think this exchange about the root cause of visible aliasing artifacts may add some insight, therefore I've prepared some examples (attached).

I wouldn't dream of contradicting your expertise in these matters if I didn't think it worthwhile for our readers, so, consider this the devil's advocate response to improve our collective understanding:

Your usual precision in these matters is not making much sense to me currently. What's inside the yellow circle is telling me precious little about aliasing. How do you propose false color as a result from what happens inside the yellow circle? What happens inside the yellow circle shows perfect anti-aliasing blur...

ALL samples are an alias. False color results from what happens directly outside the yellow circle where the aliasing is quite obvious, but very different for each channel. The difference is due to position, not density. Of course, we can go all red-herring about how the "amplitude" of the aliasing is the same for all samples, but that wasn't contested to begin with. The aliasing characteristic is the same for each sensel, the aliasing effect of sampling disjunct positions is not.
Logged
Regards,
~ O ~
If you can stomach it: pictures

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8913
Re: Fake detail, on a feather
« Reply #26 on: July 15, 2013, 11:50:18 am »

I wouldn't dream of contradicting your expertise in these matters if I didn't think it worthwhile for our readers, so, consider this the devil's advocate response to improve our collective understanding:

Your usual precision in these matters is not making much sense to me currently. What's inside the yellow circle is telling me precious little about aliasing. How do you propose false color as a result from what happens inside the yellow circle? What happens inside the yellow circle shows perfect anti-aliasing blur...

ALL samples are an alias. False color results from what happens directly outside the yellow circle where the aliasing is quite obvious, but very different for each channel. The difference is due to position, not density. Of course, we can go all red-herring about how the "amplitude" of the aliasing is the same for all samples, but that wasn't contested to begin with. The aliasing characteristic is the same for each sensel, the aliasing effect of sampling disjunct positions is not.

Hi Oscar,

By all means, we need more devil's advocates to achieve some progress, instead of the common parroting seen on forums other than LuLa.

Attached, I've included a set of RGB Raw conversion crops of the same target, one conversion in RawTherapee and one in ACR. Now it is probably more obvious that the seemingly aliased area outside the yellow circle of the individual point-samples is much smoother when the Raw converter did it's reconstruction of the [R,G1,G2,B] signals. I've also adjusted the diameter of the Yellow circle Nyquist limit to the different zoom output size with all sensels participating. Inside that circle no real signal reconstruction is possible, only aliasing exists there. Outside the circle there is also some false color aliasing, but frankly that's pretty decent given the torture test it was subjected to (and the zoom of 300%, and no artifact suppression).

Again, the earlier samples were showing the Raw sensel channels (including both the fused aliasing+signal data)  before demosaicing, hence their 'Raw looks'.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
Re: Fake detail, on a feather
« Reply #27 on: July 15, 2013, 02:10:50 pm »

Hi,

Three conversions, all monochrome, pretty much default. Raw image is here: http://echophoto.dnsalias.net/ekr/Articles/MFDJourney/FakeDetail/20130714-CF043488.iiq

Some areas marked on C1 image, but all the images show artifacts.

LR5Capture OneRaw Developer
« Last Edit: July 15, 2013, 04:43:55 pm by ErikKaffehr »
Logged
Erik Kaffehr
 

fdisilvestro

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1852
    • Frank Disilvestro
Re: Fake detail, on a feather
« Reply #28 on: July 15, 2013, 07:12:44 pm »

No, aliasing can always be removed very easily by (excessive) blurring.

The issue is that aliased signals or artifacts could be of a lower frequency than Nyquist (in theory they can be of any frequency) so, how much are you going to blur?. IMHO It is better to have a low-pass filter before sampling (not necessarily an AA layer on top of the sensor).

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8913
Re: Fake detail, on a feather
« Reply #29 on: July 15, 2013, 07:44:30 pm »

The issue is that aliased signals or artifacts could be of a lower frequency than Nyquist (in theory they can be of any frequency) so, how much are you going to blur?. IMHO It is better to have a low-pass filter before sampling (not necessarily an AA layer on top of the sensor).

Hi Francisco,

That's indeed what Erik's file demonstrates, nothing new under the sun, the image data was already aliased when it was recorded because the analog input signal was not 'properly' low-pass filtered.

To avoid such issues one can attempt to shoot additional images, e.g. with a much narrower aperture to create diffraction blur, or a bit of defocus, and then make a composite in postprocessing. That of course works best with stationary subjects. Shooting at a larger magnification factor or at an angle may also work.

Blurring the aliased image data will also destroy other useful detail, unless one uses elaborate processing tricks.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

tho_mas

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1799
Re: Fake detail, on a feather
« Reply #30 on: July 15, 2013, 07:57:55 pm »

pretty much (well, almost) no pattern moire and also no "fake" detail (as far as I can tell ...).

Processed through C1 (7.1.3) and retouched (in about 3-4 minutes) in Photoshop...

Logged

fdisilvestro

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1852
    • Frank Disilvestro
Re: Fake detail, on a feather
« Reply #31 on: July 15, 2013, 08:44:23 pm »

Much better, but even if a lot less evident, there are still artifacts or fake detail along the edges of the feather (this is splitting hairs, I know)

tho_mas

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1799
Re: Fake detail, on a feather
« Reply #32 on: July 15, 2013, 08:55:20 pm »

Much better, but even if a lot less evident, there are still artifacts or fake detail along the edges of the feather (this is splitting hairs, I know)
would require another 5-10 minutes to restore the artifact-free capture. Okay... if it's an really important image it may require half an hour or mabye even an hour (since it's "important"... who cares...).

I really appreciate technical insights to understand the limitations of certain tools.
But above all I am all for solutions how to solve issues caused by said limitations... this is why I've posted the "corrected" feather...

« Last Edit: July 15, 2013, 09:01:23 pm by tho_mas »
Logged

fdisilvestro

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1852
    • Frank Disilvestro
Re: Fake detail, on a feather
« Reply #33 on: July 15, 2013, 09:25:56 pm »

But above all I am all for solutions how to solve issues caused by said limitations... this is why I've posted the "corrected" feather...


Agreed! It is the end result that matters.

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
Re: Fake detail, on a feather
« Reply #34 on: July 16, 2013, 12:09:40 am »

Hi,

Just to make a point, I think that the image was shot at f/5.6, but I did reshoot it at f/11 and most aliasing was still there. I shot the subject with both 80/2.8 and 150/4 lenses and both had obvious aliases. Both images were shot 3.5 m, so the aliasing is much less sensitive to distance than I would have thought.

Best regards
Erik


Hi Francisco,

That's indeed what Erik's file demonstrates, nothing new under the sun, the image data was already aliased when it was recorded because the analog input signal was not 'properly' low-pass filtered.

To avoid such issues one can attempt to shoot additional images, e.g. with a much narrower aperture to create diffraction blur, or a bit of defocus, and then make a composite in postprocessing. That of course works best with stationary subjects. Shooting at a larger magnification factor or at an angle may also work.

Blurring the aliased image data will also destroy other useful detail, unless one uses elaborate processing tricks.

Cheers,
Bart
Logged
Erik Kaffehr
 

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8913
Re: Fake detail, on a feather
« Reply #35 on: July 16, 2013, 03:24:39 am »

Hi,

Just to make a point, I think that the image was shot at f/5.6, but I did reshoot it at f/11 and most aliasing was still there.

Hi Erik,

The P45+has a sensel pitch of 6.8 micron. That means that at the limiting resolution (=Nyquist if the lens is good and in the plane of best focus), it would take (2 x 0.0068) / 0.000555 = f/24.5 to reduce the MTF for 555 nanometer light to zero response as a result of diffraction. So f/22 might be enough to do the trick when there is also some residual aberration from the lens, and a tiny bit of defocus. Other wavelengths may respond a bit different, depending on the lens Chromatic corrections.

Quote
I shot the subject with both 80/2.8 and 150/4 lenses and both had obvious aliases. Both images were shot 3.5 m, so the aliasing is much less sensitive to distance than I would have thought.

When shot from the same distance, the 150mm would give higher magnification (perhaps not enough to avoid subject detail smaller than Nyquist) but also less DOF, so some shift of the focus plane would have much more impact on the COC.

When one uses a similar setup over and over again, e.g. product shots at roughly the same shooting distance with the same lens, it is relatively easy to find the correct settings for the second 'problem solving shot'. Just shoot my torture test star chart and vary the aperture and defocus amounts until you get the best compromise between moiré and detail. Then use that info for the regular 'second shot'. Figuring it out once, can save a lot of post-processing time later.

That's on the input/capture side of things. If there is still some disturbing moiré, then one can utilize the various tools in the Raw converters, from (local) moiré suppression to negative Clarity or Structure, or even use huge amounts of local noise reduction for a quick fix. Not applying CA correction can also reduce luminosity moiré a bit, but may increase False color moiré. If then there is still an issue or the Raw conversion steps took too much of a toll on image detail, Photoshop can be used to fix things, inverted High-pass filters on specific channels, or cloning from other channels to repair the offending one(s), or plain dodging and burning at the channel pixel level. Color moiré can often be corrected quite well, e.g. with a color blending mode layer, but luminosity can be a bit harder.

Of course not all subjects have such fine high contrast repetitive detail, but things like feathers and fabric and distant bricks are predictable troublemakers. The aliasing on less repetitive structures is also there, but it's just less noticeable.

It's basic Digital Signal Processing, when the detail is smaller than the regular sampling resolution (Nyquist frequency with a good lens and well focused), aliasing will always exist, unless a low-pass filter of some sort is used or the subject contrast is low enough.

Cheers,
Bart
« Last Edit: July 16, 2013, 03:42:18 am by BartvanderWolf »
Logged
== If you do what you did, you'll get what you got. ==

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
Re: Fake detail, on a feather
« Reply #36 on: July 16, 2013, 02:42:15 pm »

Hi,

I made three shots of the feather with:

P45+, 6.9 micron, no OLP
A99, 6 Micron OLP
A77, 3.9 Micron, OLP

One interesting observation is that Alpha 99 image shows far less artefacts and better detail than the P45+. The A99 has 6 micron pixels while the P45+ has 6.9 micron pixels. So the 15% resolution advantage in combination with adequate OLP filtering gives superior image quality on the pixel level for the A99. This indicate that the MFD would do well with an OLP-filter. Note that I am looking at fixed size image, so I ignore the size of the format. What I say is that for the formats I have tested, smaller pixels and OLP filter are better than larger pixels without OLP filter, according to what I see.


The images were scaled to approximately same size using "convert" from ImageMagick using default options. LuLa forums show them in reduced size, unfortunately.
P45+A99A77

Resized images here, these are the best comparison, but make sure to view at actual pixels:
http://echophoto.dnsalias.net/ekr/Articles/MFDJourney/FakeDetail/0716/20130716-CF043522PhaseOne_A_SP45+_small.jpg
http://echophoto.dnsalias.net/ekr/Articles/MFDJourney/FakeDetail/0716/20130716-_DSC2300SONYSLT-A99V_small.jpg
http://echophoto.dnsalias.net/ekr/Articles/MFDJourney/FakeDetail/0716/20130716-_DSC5172SONYSLT-A77V_small.jpg

Original crops (different size):
http://echophoto.dnsalias.net/ekr/Articles/MFDJourney/FakeDetail/0716/20130716-CF043522PhaseOne_A_SP45+.jpg
http://echophoto.dnsalias.net/ekr/Articles/MFDJourney/FakeDetail/0716/20130716-_DSC2300SONYSLT-A99V.jpg
http://echophoto.dnsalias.net/ekr/Articles/MFDJourney/FakeDetail/0716/20130716-_DSC5172SONYSLT-A77V.jpg
« Last Edit: July 17, 2013, 12:40:20 am by ErikKaffehr »
Logged
Erik Kaffehr
 
Pages: 1 [2]   Go Up